Adoption of Agricultur

World Development 146 (2021) 105599
Contents lists available at ScienceDirect
World Development
journal homepage: www.elsevier.com/locate/worlddev
Adoption of agricultural technology in the developing world:

A meta-analysis of the empirical literature
Sacha Ruzzante a,⇑, Ricardo Labarta b, Amy Bilton a
a
University of Toronto, 55 St. George St., Toronto, ON M5S 0C9, Canada
b
International Center for Tropical Agriculture (CIAT), Km 17, Recta Cali-Palmira, Valle del Cauca, Colombia
a r t i c l e i n f o a b s t r a c t
Article history: Agricultural technologies have long been promoted by governments and development organizations as
Accepted 30 May 2021 effective ways to increase farm productivity and reduce poverty. However, adoption of many seemingly
Available online 17 June 2021 beneficial technologies remains low. Empirical adoption studies attempt to identify the motivation for
adoption based on differences in characteristics between adopters and non-adopters. This study investi-
Keywords: gates variables that regularly explain adoption across technologies and contexts using a meta-analysis of
Adoption 367 regression models from the published literature. We find that, on average, farmer education, house-
Agricultural technology
hold size, land size, access to credit, land tenure, access to extension services, and organization member-
Developing world
Meta-analysis
ship positively correlate with the adoption of many agricultural technologies. Technologies in the
Global categories of improved varieties and chemical inputs are adopted more readily on larger farms, which
casts doubt on the scale-neutrality of these technologies. Agricultural credit can positively influence
adoption, but researchers should measure whether farmers are credit constrained, rather than simply
whether or not they have access to credit. While extension services may substitute for education in
the case of improved varieties, the two variables appear to be complementary for natural resource man-
agement technologies. Land tenure can encourage adoption of natural resource management techniques,
and we find it to be most influential in the adoption of technologies with long planning horizons, such as
erosion control methods. Unsurprisingly, although some patterns are identified when results are aver-
aged, most adoption determinants vary widely by technology, cultural context, and geography. Based
on these observations, we provide some recommendations for adoption researchers and policy makers,
but, given the variability of the results, conclude that efforts to promote agricultural technologies in
the developing world must be adapted to suit local agricultural and cultural contexts.
Ó 2021 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
1. Introduction to develop technologies that increase yields, reduce exposure to

environmental shocks, produce more nutritious crops, reduce
The development and dissemination of novel agricultural tech- human labour requirements, and promote long-term sustainabil-
nologies is seen as a way of enhancing productivity on the world’s ity; it is estimated that in 2011 global spending on agricultural
475 million small (<2 ha) farms, many of which are in low- and R&D totalled approximately $70 billion (2009 PPP $) (Pardey,
middle-income countries (Lowder, Skoet, & Raney, 2016). Develop- Chan-Kang, Dehmer, & Beddow, 2016).
ment experts generally agree that the responsible development of Assessments of this investment have concluded that it has been
agriculture is a key element in addressing, at least, sustainable a highly efficient use of public funding (Maredia & Anthony, 2010;
development goals 1 (no poverty), 2 (zero hunger), 3 (good health Pardey, Andrade, Hurley, Rao, & Liebenberg, 2016). However, it is
and well-being), 10 (reduced inequalities), 12 (responsible con- recognized that aggregate measures of agricultural research effec-
sumption and production), 13 (climate action), 14 (life below tiveness mask a large amount of heterogeneity: few projects drive
water), and 15 (life on land). Since the mid-20th century, the inter- the average up, while many fail (Maredia & Anthony, 2010). Wide-
national community has devoted a significant amount of resources spread adoption of a new technology is a necessary (though not
sufficient) condition for achieving impact.
Since Ryan and Gross (1943) observed that the adoption of
⇑ Corresponding author. hybrid corn occurred in a rapid but heterogeneous manner in farm-
E-mail addresses: [email protected] (S. Ruzzante), [email protected] ing communities in Iowa, researchers have studied the factors that
(R. Labarta), [email protected] (A. Bilton).
https://doi.org/10.1016/j.worlddev.2021.105599
0305-750X/Ó 2021 The Author(s). Published by Elsevier Ltd.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
S. Ruzzante, R. Labarta and A. Bilton World Development 146 (2021) 105599
lead to acceptance of agricultural innovations. One tool that has through a society. This paradigm follows from the pioneering work
been used since at least the 1970’s is the ‘adoption study,’ which of Ryan and Gross (1943), while Rogers (2003), first published in
aims to identify motivating factors and constraints to adoption 1962, remains a seminal work that defined the field of innovation
by observing differences between ‘adopters’ and ‘non-adopters’ of diffusion research. This field focuses on the characteristics of inno-
an innovation. Although hundreds of adoption studies have been vations, and how they impact rates of diffusion. Societies are
published, there have been relatively few attempts to synthesize assumed to be composed of a range ‘adopter categories’, from inno-
their findings. Previous reviews in the published literature have vators and early adopters to laggards, which differ on measurable
used relatively small sample sizes (less than 60 studies), and stud- socioeconomic, personality, and communication attributes
ies in the developing world have not used formal quantitative (Rogers, 2003). Innovation-diffusion theory has been criticized
meta-analysis. This paper aims to derive recommendations for pol- for assuming that innovations will be appropriate (Adesina &
icy makers and adoption researchers by analysing a large sample of Zinnah, 1993; Negatu & Parikh, 1999), which Rogers (2003) refers
studies across a diverse range of agricultural technologies in the to as ‘pro-innovation bias’.
developing world. Effect sizes for 15 variables are derived using The economic constraints paradigm postulates that farmers aim
meta-regression, which is a more powerful and consistent tool to maximize utility, and that uneven resource endowments lead to
for meta-analysis than the vote-count methodology used in many observed patterns of adoption (Adesina & Zinnah, 1993; Negatu &
previous studies. Parikh, 1999). In comparison to the innovation-diffusion paradigm,
the economic constraints model emphasizes the role of economic
factors at the individual level in determining adoption decisions.
2. Background However, this model allows for only strictly rational and informed
behaviour and fails to capture the effects of cultural and individual
2.1. Agricultural technology adoption perceptions of an innovation.
The adopter-perception paradigm allows for a level of subjec-
‘Agricultural technology’ is a broad term that is used here to tivity by contending that it is the perceived need to innovate and
describe equipment, genetic material, farming techniques, and the perceived attributes of innovations that determine adoption
agricultural inputs that have been developed to improve the effec- behaviour (Kivlin & Fliegel, 1967; Adesina & Zinnah, 1993). Charac-
tiveness of agriculture. ‘‘Effectiveness” refers to an equally broad teristics of an innovation and its delivery combine with cultural,
range of productivity, health, welfare, and sustainability outcomes. contextual, and individual factors to influence perception
The agricultural technologies surveyed here can be categorized (Adesina & Zinnah, 1993; Prager & Posthumus, 2010). Meijer,
into four broad categories, which are presented in Table 1 along Catacutan, Ajayi, Sileshi, and Nieuwenhuis (2015) present an ana-
with some representative examples. lytical model of decision making that includes the intrinsic factors
Agricultural technology adoption theory is a multi-disciplinary of knowledge, perceptions, and attitudes, which are conditioned by
field that combines elements of decision theory and diffusion of extrinsic factors, such as characteristics of the farmer, the external
innovations theory in an attempt to explain why some farmers environment, and the innovation. Within the adopter-perception
adopt new technologies and others do not. The literature falls into paradigm, farmers may still be considered rational actors who
three paradigms: the innovation-diffusion paradigm; the economic maximize utility; however, in contrast to the economic constraints
constraints paradigm; and the adopter-perception paradigm paradigm, the definition of utility is expanded beyond simple
(Adesina & Zinnah, 1993; Prager & Posthumus, 2010). Each para- financial considerations.
digm emphasizes the role of different factors on adoption rates The influence of many factors on adoption can be explained by
and patterns. any of the three paradigms. For example, the innovation-diffusion
The innovation-diffusion paradigm assumes that information is paradigm holds that earlier adopters tend to have larger farms than
the critical parameter that controls the spread of an innovation later adopters. However, this paradigm does not distinguish
Table 1
The technologies examined in this study are comprised of four major categories, shown here with some representative examples.
Description Examples
Natural Resource Management Technologies that have been ‘‘developed to deal Minimum/zero tillage (Teklewold et al., 2013)
with and mitigate environmental stresses, such as Permanent soil cover (Wainaina et al., 2016)
land degradation and nutrient depletion” (Wainaina Intercropping (Dalton et al., 2011)
et al., 2016). Crop rotation (Ndiritu et al., 2014)
Rock & soil bunds Gebremedhin and Swinton (2003)
Organic farming (Sodjinou et al., 2016)
Terracing (Amsalu & de Graaff, 2007)
Agroforestry (Neupane et al., 2002)
Improved Varieties New crop varieties that are developed to exhibit High-yielding maize (Mukasa, 2018)
favourable characteristics, such as increased yields, Pigeonpea developed to increase food and fuel supply (Orr et al., 2015)
pest resistance, drought tolerance, shorter growing Short duration mangrove swamp rice varieties (Adesina & Zinnah, 1993)
periods, higher nutritional value, etc.
Disease-resistant lentils (Yigezu et al., 2019)
Chemical Inputs Agrochemical inputs intended to increase yields or Pesticides (Mukasa, 2018)
reduce losses to pests Inorganic fertilizer (Ogada et al., 2014)
Mechanization and Infrastructure Any of a variety of technologies requiring Petrol/diesel groundwater pump (Gebregziabher et al., 2014)
significant investment in physical equipment. Rice and wheat thresher (Mottaleb et al., 2016)
Treadle pump (Adeoti, 2008)
Grain storage silo (Bokusheva et al., 2012)
Laser-land levelling (Ali et al., 2018)
2
whether early adoption of innovations is the cause or result of higher yields, which increases the potential profit from adopting
increased wealth and resources (Rogers, 2003). The economic con- improved varieties of seeds. They may also influence expected util-
straints paradigm might argue that larger farms are better able to ity indirectly, through a relationship with risk. For example, educa-
exploit economies of scale required to make adoption of an innovation may reduce risk-aversion, and thus encourage more educated
tion profitable. The adopter-perception paradigm would reason farmers to adopt innovations, which promise higher incomes but
land size correlates with the perceived need to innovate, as well less security (Knight, Weir, & Woldehanna, 2003).
as with positive perception of the innovation. Many empirical Thus, it is assumed that adoption is the realized value of an
adoption studies construct econometric models which combine unobserved latent utility function U. U is the estimate of U, which
variables derived from all three paradigms (Adesina & Zinnah, is most often assumed to be a linear function of X (Marenya &
1993; Negatu & Parikh, 1999; Ntshangase, Muroyiwa, & Sibanda, Barrett, 2007):
2018). In this study, adoption theories are used to interpret the
U ¼ Xb þ e ð2Þ
strength and direction of correlations between predictor variables
and adoption outcomes. Where b is a vector of estimated parameters and e is a random error
th
term. The i farmer is assumed to adopt if the expected utility of
2.2. Culture and context adopting the innovation is greater than 0:

The role of culture and context in adoption is widely recognized 1; if U i P 0
yi ¼ ð3Þ
and is increasingly explicitly considered in the empirical adoption 0; otherwise
literature. Andersson and D’Souza (2014), for example, situate the
where yi is the observed adoption behaviour of the farmer. Many
promotion and adoption of conservation agriculture practices
variations of this basic model are found in the literature. For exam-
throughout southern Africa within the political and socioeconomic
ple, some authors choose to measure intensity of adoption (e.g.
contexts of Malawi, Zambia, and Zimbabwe in the preceding dec-
Kristjanson, Okike, Tarawali, Singh, & Manyong (2005) measure
ades. Wainaina, Tongruksawattana, and Qaim (2016) investigate
the ratio of land planted with improved dual-purpose cowpea to
the role that infrastructure and institutional variables, such as dis-
the total area planted with all varieties of cowpea). In this case,
tance to market and access to information, as well as geographical
the outcome is often assumed to be a censored linear function of
variables, such as drought and flooding frequency and agroecolog-
expected utility:
ical zone, play in the adoption of various agricultural technologies

across Kenya. Matuschke and Qaim (2009) investigate the role of cU i ; if U P 0
caste and social networks in the adoption of hybrid wheat and yðintensityÞi ¼ ð4Þ
0; otherwise
pearl millet in Maharashtra, India. Larsen (2018) describes a ‘soli-
darity chain’ principle that was implemented in an agricultural where c is a scaling term.
and livestock project in the Arusha region of Tanzania, which influ- Other innovative approaches that have appeared recently in the
enced the adoption of improved banana plants among non-project literature include duration analysis to investigate the timing of
farmers. While cultural and contextual factors clearly do affect adoption (e.g. Dadi, Burton, & Ozanne, 2004; Yigezu et al., 2018),
adoption patterns, such factors are difficult to incorporate into a multi-hurdle regressions to assess various stages of, or constraints
quantitative meta-analysis. An in-depth qualitative investigation to, adoption (e.g. Lambrecht, Vanlauwe, Merckx, & Maertens, 2014;
of culture and context in the meta-analysed studies is beyond Shiferaw, Kebede, Kassie, & Fisher, 2015), ordinal dependent vari-
the scope of the current work. ables to describe extent of adoption (e.g. Wollni, Lee, & Thies,
2010; Teklewold, Kassie, & Shiferaw, 2013), multinomial regres-
2.3. Adoption studies sions to analyse choices between more than 2 technologies, or
between specific combinations of technologies (e.g Fisher, Abate,
‘‘Adoption studies” of agricultural innovations have been car- Alemayehu, & Madulu, 2015; Khonje, Manda, Mkandawire, Tufa,
ried out since at least the 1970’s. These studies use correlational & Alene, 2018), and the explicit consideration of disadoption as
techniques to explain adoption behaviour as a function of variables separate from non-adoption (e.g. Marenya & Barrett, 2007;
that are gathered through farmer surveys. In other words, they aim Grabowski, Kerr, Haggblade, & Kabwe, 2016). Despite the wide
to estimate a function of the form: variety of dependent variable definitions and estimation tech-
niques, all adoption studies rely on the underlying theory of utility
Adoption ¼ f ðXÞ ð1Þ
maximization. This theoretical consistency is a necessary condition
where ‘adoption’ is the observed adoption behaviour of the farmer to make meta-analysis a valid exercise.
and X ¼ fx1 ; x2 ; . . . xn g is a matrix of socioeconomic, personality, Adoption studies which follow the basic structure outlined
environmental, farm financial, farm management, or external fac- above are subject to a number of limitations, which have been dis-
tors, which are gathered through surveys. The most common vari- cussed by various authors, most notably Doss (2006) and Burton
ables are provided in Table 3. It should be noted that some (2014), and which are briefly summarised in the following sec-
studies have also begun to include ‘perception’ variables in the tions. Some of these limitations can bias regression estimates,
matrix X, which follow from the adopter-perception paradigm. which translates to bias in the present meta-analysis. Fortunately,
For example, Amsalu and de Graaff (2007) include farmers’ percep- statistical remedies do exist to rectify these biases. Throughout this
tions of tenure security, Njuki, Mapila, Zingore, and Delve (2008) meta-analysis special attention is paid to studies that use more
include perceptions of poverty, fertilizers, and soil fertility, and advanced statistical methodologies to discern and correct for these
Teshome, de Graaff, and Kassie (2016) include perceptions of soil limitations.
erosion problems and of soil and water conservation techniques.
However, these variables do not yet appear frequently enough in 2.3.1. Lack of dynamics
the literature to warrant inclusion in this meta-analysis. Adoption is a dynamic process, which can include a trial stage,
Adoption studies assume that farmers are rational actors who early/late adoption, partial adoption, and disadoption. This infor-
aim to maximize an unobserved expected utility function mation is not usually found in cross-sectional data. Understanding
(Adesina & Zinnah, 1993). Variables may influence expected utility these dynamics is an important component of understanding the
directly, as in the case of soil fertility: higher soil fertility leads to decisions made by individual farm households and is particularly
3
useful for drawing causal inferences from correlations. As Doss However, since gender is expected to affect adoption through land
(2006) notes, studies that rely on cross-sectional data cannot, in tenure, the effect of land tenure will be underestimated. Defining
general, distinguish whether predictor variables are the cause or explicit hypotheses, through causal diagrams or other methods,
effect of technology adoption (or both or neither). Doss calls for will in general lead to better-specified models and more inter-
more studies to employ panel data (where the same individuals pretable and consistent results. Throughout the present study,
are surveyed at several points in time) to address this, which sev- results from models that were obviously biased by covariant pre-
eral authors have done (Ogada, Mwabu, & Muchai, 2014; Mukasa, dictors were removed1.
2018; Katengeza, Holden, & Lunduka, 2018; Arslan, Belotti, &
Lipper, 2017; Verkaart, Munyua, Mausch, & Michler, 2017; 2.3.4. Non-exposure bias
Wossen, Berger, & Di Falco, 2015). Others have constructed Diagne and Demont (2007) show that if only part of the popu-
quasi-panel data based on farmer recall (Moser & Barrett, 2006). lation is aware of a technology, estimating a classical adoption
However, many of these studies use the panel nature of their data model (Eq. 3) leads to biased estimates. This occurs because aware-
to control for unobserved heterogeneity via Mundlak-Chamberlain ness of the technology is not randomly distributed within the pop-
fixed effects (Mundlak, 1978), rather than to explicitly establish ulation. Several authors have since used similar Average Treatment
causality (via lagged variables). Effect (e.g. Kabunga, Dubois, & Qaim, 2012), Heckman (e.g.
In the absence of panel data, some authors have used instru- Lambrecht et al., 2014), or other multi-hurdle regression models
mental variables to address endogeneity (Mauceri, Alwang, (e.g. Shiferaw et al., 2015) to account for incomplete exposure to
Norton, & Barrera, 2007; Di Falco & Bulte, 2013; Kristjanson a technology. In the context of this meta-analysis, consistent biases
et al., 2005). However, it is generally impractical or impossible to were not detected by comparisons between studies that corrected
find valid instruments for all potentially endogenous variables. for exposure and studies that did not.
These studies, which represent a higher level of statistical care
than many other studies included in the present meta-analysis,
2.3.5. Pro-innovation bias
provide valuable robustness checks. Confidence in the meta-
As noted by Doss (2006), adoption studies tend to implicitly
analysis is increased where these studies report results that are
assume that the technology being studied is in fact better than
consistent with the larger data set.
other options available to the farmer. In some cases, farmers may
maximize utility by choosing not to adopt, choosing to delay adop-
2.3.2. Lack of variability within samples
tion, or choosing to partially adopt technologies. Baumgartner and
Doss (2006) comments that micro-adoption studies that exam-
Cherlet (2016) make an important distinction between privately
ine adoption within a single cultural, agro-ecological, and policy
and socially optimal rates of adoption of soil conservation prac-
environment cannot address important questions about how pol-
tices; the private rate of adoption is unlikely to reflect what would
icy makers should promote adoption. A number of national-level
be optimal from a societal or environmental perspective.
and international studies have been conducted since 2006
Even for new technologies that clearly outperform existing
(Mukasa, 2018; Kassie, Teklewold, Jaleta, Marenya, & Erenstein,
technologies, non-adoption may be controlled by factors that are
2015; Arslan et al., 2017; Dalton, Lilja, Johnson, & Howeler, 2011;
unrelated to a farmer’s desire to adopt. In this respect, Andersson
Fisher et al., 2015; Wainaina et al., 2016) which encompass vari-
and D’Souza (2014) differentiate between constraints and prereq-
ability in government policies and agro-ecological variables, and
uisites to adoption. While constraints may include factors that vary
have correlated this variability with adoption outcomes. Since
by farm, such as lack of financial capital, labour availability, low
the number of such studies remains small, a quantitative meta-
soil fertility etc., prerequisites refer to wider institutional, eco-
analysis is not yet justified, and is out of the scope of the current
nomic, and social conditions that must be in place to enable adop-
work.
tion. If the prerequisite conditions for adoption, such as input
availability and functional markets, are not met, then the rational
2.3.3. Model misspecification
farmer will choose not to adopt a new technology. Some adoption
Many adoption studies contain misspecified models, which
studies have analysed prerequisite conditions at the individual
either include covariant predictors that explain overlapping por-
level, such as awareness of a technology (non-exposure bias, as
tions of the variance in adoption, or exclude important predictors.
above), access to inputs, and access to credit (Shiferaw, Kebede,
Both problems lead to biased coefficient estimates, although the
& You, 2008; Shiferaw et al., 2015).
first problem (covariant predictors) usually leads to underesti-
mated coefficients, while the second (omitted variables) is more
likely to lead to overestimated coefficients. These misspecifications 2.3.6. Varietal identification
stem from a failure to make appropriate connections to adoption As with all survey-based research, sampling and survey
theories; as a result, many adoption studies provide little informa- methodology can bias results in unpredictable ways. A salient
tion on causal relationships (Burton, 2014). example is given by Floro, Labarta, Becerra López-Lavalle,
One example of this is the lack of distinction between root and Martinez, and Ovalle (2018), who use a DNA fingerprinting
proximate factors. Root factors are entirely independent, and are approach to show that relying on farmers to self-identify the vari-
usually believed to affect utility indirectly through relationships eties of cassava they are growing leads to inaccurate assessments
with proximate factors. Proximate factors are those for which a of overall adoption rates, as well as biased predictor estimates.
direct causal link to the utility function can be drawn. For example, Yigezu, Alwang, Rahman, and Mollah (2019) find similar results
gender, a root factor, may affect the utility of adoption through with improved lentil varieties. More work is needed to understand
land tenure, a proximate factor. Women may be less likely to the prevalence of variety misidentification and its effect on adop-
own their land, which may discourage them from adopting tech- tion study results, to say nothing of its effect on farm productivity.
nologies with long-term benefits.
1
Including root and proximate factors in the same multiple For example, where farm size and farm size squared were included in the same
regression model will lead to biased and inconsistent estimates. regression, and where the marginal effect of farm size squared was not much smaller
than the marginal effect of farm size. There may be legitimate reasons to include
If land tenure and gender are included in the same model, the coef- squared variables, but each coefficient cannot then be interpreted in isolation. We
ficient of land tenure represents the residual effect of land tenure also note that age and experience are also often highly covariant; we removed results
on adoption after partialling out variance correlated with gender. where both these variables were included in the same regression.
4
2.4. Previous syntheses of adoption research opment index (HDI) of 0.7 or less were included. For studies that
did not use nationally representative data, the HDI was computed
There have been numerous previous attempts to synthesize as the average of the subnational HDI values in the regions where
empirical adoption research. Feder, Just, and Zilberman (1985) data were collected (Smits & Permanyer, 2019). The data collection
conducted a systematic review of the adoption literature, which and cleaning are fully described in Ruzzante and Bilton (2021).
remains a highly valuable resource for researchers interested in Here, the focus is on four technology categories that comprise
classical theories of adoption. However, the empirical adoption lit- 337 out of the 367 regressions (Table 1).
erature has grown tremendously since 1985, both in number of
studies published and in the use of advanced econometric 3.2. Data preparation
methodologies.
Table 2 summarizes previous syntheses of agricultural technol- Several data transformations are used to compare results from
ogy adoption studies, beginning with Feder et al. (1985). Authors studies that employ different statistical models and reporting
have primarily used narrative reviews and vote-counting proce- styles.2 Two comparison metrics are used: the partial correlation
dures to synthesize the published empirical research. While narra- coefficient (PCC), and the Log-Odds Ratio. Partial correlation coeffi-
tive reviews can provide valuable insights, they are highly subject cients can be computed from a wide range of regression models
to author biases, and lack the ability to quantify effect sizes and and can be used to compare the strength of a relationship between
sources of error (Stanley, 2001). Vote-counting, where the number studies that use different dependent and independent variable defi-
of positive, negative, and non-significant results are tallied, is an nitions (Ogundari & Bolarinwa, 2018), whereas the log-odds ratio
inefficient and inconsistent technique, with power that decays can only be compared for Logit or Probit models that use identical
asymptotically to zero for the combination of effect sizes and sam- or scaled dependent variables. In the present meta-analysis, 36 of
ple sizes often found in social science research (Hedges & Olkin, 204 primary studies use non-binary dependent variable definitions.
1980; Friedman, 2001). While consistent tests have been devel- Independent variables are also inconsistently defined in the litera-
oped for vote-counting analyses, these are only recommended for ture. Therefore, there is usually more data available for the PCC anal-
studies which do not report enough information to calculate an ysis than for the Log-Odds Ratio analysis.
effect size (Bushman & Wang, 1995). In the present study, effect The squared partial correlation coefficient represents the pro-
sizes were calculated for 193 out of 204 studies. portion of variance in the dependent variable that is uniquely asso-
Baumgart-Getz, Stalker, and Floress (2012) use mixed-effects ciated with the independent variable. However, it excludes any
ANOVA with a sample of 46 studies of ‘best management practices’ variance in the dependent variable which could be explained by
in the United States. This quantitative meta-analysis allows them two or more independent variables. Cohen (1988) suggests that
to weight the influence of each study based on its reported preci- partial correlations of 0.1, 0.3, 0.5 correspond to small, medium,
sion, as well as control for heterogeneity in the results caused by and large effects. Doucouliagos (2011) recommends more liberal
differences in study methodology and region. Rubas (2004) pre- guidelines of 0.07, 0.17, and 0.33 for small, medium, and large
sented a regression analysis using 107 studies that investigated a effects.
range of agricultural and non-agricultural technologies from
around the world (including developed countries). Unfortunately,
3.3. Meta-regression
much of the analysis is based on the common error of assuming
that a non-statistically significant result is evidence of no effect
Meta-regression is a method of weighted least-squares regres-
(Ranganathan, Pramesh, & Buyse, 2015).
sion that accounts for within-study sampling variance. It is used
Despite the amount of work done, there has not been a large
to compare effect sizes based on magnitude and precision. Random
(> 100 studies) meta-analysis of agricultural technologies for glo-
or mixed-effects meta-regressions allow for the true effect size to
bal development, and the range of technologies considered in pre-
vary due to heterogeneity in observed variables such as geography
vious meta-analyses has generally been limited. To the best of the
and type of technology, as well as unobserved variables. The pre-
authors’ knowledge, this study presents a meta-analysis of the lar-
dicted values of random-effects meta-regression can be inter-
gest dataset of agricultural technology adoption studies to date. In
preted as the mean effect size across studies. The weight for the
addition, this article presents the first truly quantitative meta- th
analysis (using meta-regression) of such studies focused on the i estimate wi is:
developing world. The size and diversity of the dataset and the 1
use of meta-regression enable an analysis of adoption determi- wj ¼ ð5Þ
ðs
^2 þ s2j Þ
nants that is less prone to author bias than previous narrative
reviews and more statistically powerful than previous analyses.
where s2j is the estimate of the sampling variance r2j of the j study
th
Meta-regression also allows for meaningful comparisons of adop-
tion determinants for different categories of technology, as well and s^2 is an estimate of the inter-study heterogeneity s2 . There are
as investigation of methodological and contextual factors that many methods in the literature on how to estimate s ^2 . Paule and
may affect study results. Mandel (1982) describe an iterative method that is robust to devi-
ations from normality, recommended by DerSimonian and Kacker
(2007), which is used here. The R package metafor is used
3. Methodology (Viechtbauer, 2010), which offers 7 distinct s^2 estimators. A sensi-
tivity analysis has demonstrated that although s
3
^2 is sensitive to
3.1. Data collection the estimator used, the meta-regression results are highly
insensitive.
A convenience sample of 204 adoption studies was collected As a first estimate of predictor variable effect sizes, meta-
from peer-reviewed and grey literature. Many studies include regression is performed without moderating variables for each pre-
more than one adoption model, resulting in 367 distinct regression dictor variable in Table 3 and each technology category in Table 1.
models. The models track adoption of agricultural technologies in
43 countries across five continents (see Fig. 1). Studies of adoption 2
See Supplemental Appendix Table A1.
in regions that, at the time of data collection, had a human devel- 3
See Supplemental Appendix Figs. A1 and A2.
5
Table 2
Previous syntheses of agricultural technology adoption studies.
Authors Technology Region Methods Sample Notes

Size
Feder et al. General Developing Narrative – Seminal review which established many classical theories of adoption.
(1985) World review
Pattanayak Agroforestry Primarily Vote- 32 Conclude that adoption is most often correlated with variables that measure risk, bio-
et al. (2003) Tropics count physical conditions, and resource-endowment.
Rubas (2004) General Global Regression 107 Thesis dissertation; Includes some non-agricultural technologies.
Knowler and Conservation N. America, L. Vote- 23 Conclude that ‘‘ there are few if any universal variables that regularly explain the
Bradshaw Agriculture America, & count adoption of conservation agriculture across past analyses.”
(2007) Africa
Baumgart-Getz Best United States Mixed- 46 Conclude that access to information, financial capacity, and being connected to agency
et al. (2012) Management effects or networks of farmers had the largest impact of technology adoption.
Practices ANOVA
Wauters and Conservation Developed Vote- 38 Conclude that there are ‘‘almost no universal patterns to discover.”
Mathijs Agriculture Countries count
(2014)
Burton (2014) Environmental Developed Narrative 53 Concludes that inconsistencies in the published literature are mainly due to a
Behaviour Countries review widespread failure to meaningfully investigate causality.
Liu et al. (2018) Best Global Narrative 100 Suggest future research should focus on study scale, on measuring adoption as a
Management review continuous process, and on the role of social norms and uncertainty in decision-
Practices making.
Fig. 1. The geographic distribution of the dataset of regression models reported in adoption studies in the present meta-analysis.
This is equivalent to a weighted mean. Throughout this study ‘‘sig- tor of moderators (a subset of Table 4), cij is a vector of estimated
nificance” is defined as 95% confidence, unless otherwise specified. coefficients, and eij is an error term.
Because of violations of normality in much of the data presented
here, ordinary non-parametric bootstrapping with 10,000 replica- 3.4. Limitations
tions is used for all meta-regressions without moderators with
more than 4 observations. This study is limited by many of the problems that plague adop-
Methodological, regional, and other factors may affect the tion studies in general (discussed in Section 2.3). In addition, meta-
results found in adoption studies. Therefore, study-level modera- analyses impose their own limitations. These are summarized
tors are coded and included in the meta-regression. The meta- below.
regression takes the form:
3.4.1. Non-random sampling
Y ij ¼ cij Mij þ eij ð6Þ In this work, studies are not randomly sampled from all tech-
nology adoption contexts. Easily accessible papers, from recent
years, published in better-quality journals are more likely to be
where Y ij is the estimated expected value of the i
th
predictor vari- included. Even if every study ever conducted were to be included,
th
able (Table 3) for the j technology grouping (Table 1), Mij is a vec- it is unlikely that the sample would be representative of all regions
6
Table 3 greater than 0.1, Cohen’s guideline for a ‘‘small” effect size. Nine
Standard predictor variable definitions for the 15 most common variables included in of 57 estimates pass Doucouliagos’ more liberal guideline of 0.07.
studies.
The reader is reminded that the confidence intervals shown in
Farmer/farm household characteristics Figs. 2 and 3 express confidence in the mean effect size, rather than
Age Farmer’s age (years) an estimate of the variability found in the data. Although few vari-
Education Farmer’s years of formal education (years)
Gender Farmer’s gender, male = 1
ables are found to be consistently strong predictors of adoption,
Household Size Number of people living in household as principal many studies do find large, but conflicting, effect sizes4. The I2
residence statistic is an estimate of the ratio of residual heterogeneity to total
Experience Number of years of farming experience (years) variability in a meta-regression (Higgins & Thompson, 2002). Higher
Farm Biophysical characteristics values indicate that more variability in the results is due to real dif-
Land Size Size of farmer’s landholding (ha) ferences between studies, rather than sampling error. In the average
Fertility Soil fertility (highly fertile = 1)1
results presented in Fig. 2 and 3, the median value of I2 is 84%. The
Slope Slope of land (steep = 1) 2
conclusion is that factors that influence adoption are heterogeneous
Financial characteristics
across different contexts and technologies. The large sample sizes
Dist Distance to market or other relevant source of inputs/
outputs (km) included in this study are therefore crucial in order to provide accu-
Credit Farmer has access to credit (true = 1) rate mean effect sizes, as well as to investigate sources of
Land Tenure Farmer has secure tenure of land (true = 1) heterogeneity.
Livestock Livestock owned by the farmer (Tropical Livestock Knowler and Bradshaw (2007) found similar results in their
Units)
Non-Farm Income Farmer has non-farm sources of income (true = 1)
vote-counting analysis of conservation agriculture studies, con-
cluding that there are ‘‘few if any universal variables that regularly
External factors
Extension Farmer has contact with extension workers (true = 1) 3
explain the adoption of conservation agriculture across past analy-
Organization Farmer is a member of a farmer ses”. Wauters and Mathijs (2014) state that there are ‘‘almost no
Membership association/cooperative (true = 1) universal patterns to discover” in the adoption literature on soil
1
Soil fertility is somewhat subjective and region-dependent. Unfortunately
conservation practices in the developed world. It appears that
details about how fertility was measured, and what constitutes fertile soil, are these statements apply to a broad range of agricultural technolo-
infrequently reported. gies, and not just conservation agriculture.
2
What constitutes a steep slope is also region-dependent. While some studies This discussion will focus on seven predictors which are often
use numeric cut-offs for steepness (Teshome et al., 2016), others rely on farmer or
assumed to positively influence adoption: Education, Household
surveyer opinion (Kassie et al., 2015).
3
While most studies include contact with any agricultural extension service, Size, Land Size, Access to Credit, Land Tenure, Access to Extension,
some studies specify contact with specific programs (e.g. Diagne and Demont, 2007; and Organization Membership. For each predictor, mean results
Katengeza et al., 2015), or extension services related to particular technologies (e.g. will be compared with previous meta-analyses and with select
Ojiako et al., 2007; Yigezu et al., 2018). studies that use advanced econometric tools to provide more reli-
able results. Explanations will be offered by drawing on adoption
theories. Where appropriate, sources of heterogeneity in the
in the developing world. Therefore, the results presented here are results will be investigated by adding moderators to the meta-
not strictly generalizable to all contexts; rather, they represent regressions.
the collective findings of adoption researchers who have made Education is significant and positively related to adoption for
their work accessible to the scientific community. all but NRM technologies. These results are consistent with Rogers’
assertion that ‘‘earlier adopters have more years of formal educa-
tion than do later adopters” (Rogers, 2003), as well as Feder et al.
3.4.2. Lack of information on local context (1985), who conclude that ‘‘farmers with better education are ear-
Cultural, political, and policy environments are likely to influ- lier adopters of modern technologies”. Knight et al. (2003), using
ence the characteristics that drive adoption. For example, land size data from Ethiopia, find that schooling positively influences adop-
is likely to constrain adoption in regions where land ownership is tion of modern inputs both directly and indirectly, through a
highly individualized; however, many communities have access to reduction in risk-aversion. Previous meta-analyses of the effect of
large amounts of communal pasture or cropland, which can relax education on NRM technology adoption have found similarly
this constraint. However, few studies include the level of detail ambiguous results (Pattanayak, Mercer, Sills, & Yang, 2003;
required to permit quantitative meta-analysis. Thus, this study is Knowler & Bradshaw, 2007; Baumgart-Getz et al., 2012; Wauters
restricted to the simple moderators presented in Table 4. & Mathijs, 2014).
The small and non-significant effect of education for NRM
(PCC ¼ 0:007 0:011) is unexpected. Education has long been
4. Results and discussion hypothesized to influence adoption through increasing a farmer’s
ability to ‘‘perceive, interpret, and respond to new events in the
Fig. 2 graphically displays the estimated average partial correla- context of risk” (Schultz, 1981). NRM technologies are compara-
tion coefficient for the 8 predictor variables in the categories of tively complicated and risky investments, so education was
Farmer and Farm Household and Land and Biophysical Character- expected to strongly encourage their adoption. However, NRM
istics (Table 3), for each of the four technology categories. Fig. 3 technologies often require large investments in time and labour,
shows the corresponding data for Financial Characteristics and while other technologies are primarily capital investments. More
External Factors. Each estimate and confidence interval is derived educated farmers, who are often also wealthier (Rapsomanikis,
from 10,000 bootstrap samples of a meta-regression with no mod- 2015), may have more resources to invest in improved varieties,
erators. A comparison of the results using the partial correlation chemical inputs, or mechanization. However, at least in the
coefficient and the odds ratio yielded similar conclusions (shown
in Supplementary information). Given that the PCC is available
from more studies, it is used throughout the rest of this study.
Small average effect sizes are found for all predictors across all 4
See Supplemental Appendix Figs. A5 and A6 for boxplot distributions of the data,
technology categories. Only 2 of 57 expected values of PCC are which illustrate the variability in reported effect sizes.
7
Table 4
Summary statistics for regression models included in this meta-analysis. Standard deviations are provided for continuous variables.
All (N = 367) IV (N = 143) NRM (N = 93) Chem (N = 45) Mech (N = 56)

mean SD mean SD mean SD mean SD mean SD
Region:
Eastern Africa 0.41 0.39 0.56 0.53 0.23
Southern Africa 0.19 0.20 0.18 0.16 0.14
Western Africa 0.20 0.20 0.10 0.18 0.30
South Asia 0.09 0.11 0.08 0.00 0.16
South-East Asia 0.04 0.02 0.06 0.07 0.04
East Asia 0.01 0.00 0.01 0.00 0.09
Latin America 0.04 0.06 0.01 0.07 0.04
Mid. East & East. Europe 0.01 0.01 0.01 0.00 0.00
Adoption level (%) 40 (24) 46 (24) 32 (20) 38 (27) 34 (25)
Average farm size (ha) 3.1 (4.5) 3.7 (6.3) 2.5 (2.7) 2.4 (2.2) 3.0 (2.4)
Average education (years) 5.54 (2.10) 5.57 (1.97) 5.01 (2.20) 5.41 (1.83) 6.89 (1.52)
Human Development Index 0.48 (0.10) 0.47 (0.09) 0.47 (0.09) 0.46 (0.12) 0.52 (0.10)
Education Index 0.41 (0.12) 0.40 (0.10) 0.41 (0.12) 0.40 (0.14) 0.45 (0.12)
Random sampling used (1/0) 0.77 0.84 0.77 0.78 0.62
Peer-reviewed journal article (1/0) 0.74 0.67 0.82 0.64 0.77
Number of observations1 8034 (7E4) 1415 (4814) 1543 (1570) 2061 (2531) 44509 (2E6)
Number of predictors in regression 17 (8) 16 (7) 22 (8) 17 (8) 13 (5)
Plot-level observations used2 (1/0) 0.22 0.12 0.48 0.29 0.07
Multivariate regression (1/0) 0.31 0.20 0.56 0.53 0.11
Exposure correction3 (1/0) 0.08 0.15 0.00 0.02 0.07
Study geographic size:
Village 0.02 0.00 0.00 0.09 0.05
Sub-regional 0.37 0.46 0.32 0.07 0.39
Regional 0.11 0.13 0.13 0.04 0.07
Sub-national 0.31 0.28 0.32 0.36 0.27
National 0.16 0.10 0.16 0.40 0.16
International 0.04 0.03 0.06 0.04 0.04
NRM technologies only
Zero- and minimum-tillage (1/0) 0.16
Erosion control techniques4(1/0) 0.31
Fertility management5 (1/0) 0.43
Organic farming (1/0) 0.03
Various techniques (1/0) 0.06
Chemical inputs only
Fertilizer (1/0) 0.76
Pesticides and herbicides (1/0) 0.24
Mechanization and Infrastructure only
Irrigation (1/0) 0.29
Field mechanization (1/0) 0.41
Off-field mechanization/infrastructure (1/0) 0.30
1
These data are highly skewed by one study (three regression models) which used census data from Bangladesh with 814,048 observations (Mottaleb et al., 2016).
2
As opposed to household-level observations.
3
Average treatment effects, Heckman Probit, other multi-hurdle regressions, etc.
4
Countour hedgerows, soil/rock bunding, terracing, etc.
5
Intercropping, crop rotation, manure, mulching, etc.
absence of functioning labour markets, they may be no less labour- statistically significant and there was no evidence of non-
constrained than less educated farmers. linearities in the normalized effect of education. Regressions are
It was expected that the effect of education on adoption would included in the Supplemental appendix.
be non-linear. The dataset includes a wide range of education Household size is positive and significant in the adoption of
levels, from very low (e.g a rate of formal education of only 22% improved varieties, and positive and marginally significant for
among rice farmers in Côte d’Ivoire in 1992–1993 (Adesina, NRM practices (PCC ¼ 0:010 0:011), but not in the adoption of
1996)) to quite high (e.g. averages of 11 and 13 years of schooling chemical inputs or mechanization and infrastructure, which is con-
in Benue State and Federal Capital Territory, Nigeria (Alabi, Lawal, sistent with expectations from adoption theory. Household size is
Coker, & Awoyinka, 2014; Audu & Aye, 2014)). Three model spec- often considered a proxy for labour availability, which Feder
ifications were attempted; first, the PCC of Education was et al. (1985) suggest is related to adoption of labour-intensive tech-
regressed on the subnational Education Index (Smits & nologies (such as NRM practices and, to some extent, improved
Permanyer, 2019). Second, the PCC was regressed on average years varieties). This is supported by Grabowski et al. (2016), who use
of schooling, using only studies that report this information. Lastly, quantitative and qualitative data from Zambian smallholders to
the PCC was regressed on a categorical variable identifying the show that labour availability is the primary constraint in adoption
variable definition used for education. Studies in regions with of labour-intensive hand-hoe planting basins, while capital con-
low levels of education are more likely to use a binary definition, straints limit the use of the more expensive ox-ripper.
so it was reasoned that the variable definition would act as an Land size is one of the most commonly investigated factors
acceptable proxy for low education levels. None of the results were across all four technology categories, and has strong theoretical
8
Fig. 2. The bootstrapped expected values of PCC for predictors in the categories of Farmer and Farm Household and Land and Biophysical Characteristics. The expected value
of the effect of each predictor on adoption of Natural Resource Management (green), Improved Variety (orange), Chemical Inputs (purple) and Mechanization and
Infrastructure (magenta) adoption is shown as a solid square with error bars indicating 95% confidence intervals. The numbers inside the squares indicate the number of
studies used in the meta-regression. Non-significant results are dotted. (For interpretation of the references to colour in this figure legend, the reader is referred to the web
version of this article.)
roots. Feder (1982) showed that in the absence of fixed transaction The significant effect sizes for improved varieties and chemical
costs, the adoption of scale-neutral technologies (such as chemical inputs (which are supposed to be scale-neutral) suggest that there
inputs or improved varieties) will be insensitive to land size. How- may be fixed costs associated with these technologies (Feder et al.,
ever, ‘lumpy’ (indivisible) technologies, such as agricultural 1985). These could include transport costs, costs associated with
machinery, will be adopted only by farmers with a minimum crit- the knowledge acquisition needed to plant improved varieties, or
ical farm size that makes adoption profitable. Here, land size is the cost of complementary inputs or equipment needed to plant
found to have a small but significant effect on the adoption of new varieties. In addition, many improved varieties only result in
improved varieties (PCC ¼ 0:036 0:032) and chemical inputs higher yields when paired with indivisible technologies, such as
(PCC ¼ 0:029 0:029). The effect size for NRM technologies is neg- irrigation; Clay (1975) and Feder (1982) showed that, in the
ative but not significant. The effect size for mechanization and absence of abundant credit, this complementarity will impede
infrastructure is the largest mean effect size found in this study the adoption of improved varieties on small farms.
(PCC ¼ 0:131 0:074), which is consistent with expectations from The non-scale-neutrality of improved varieties is supported by
theory. Fischer (2016), who provides a critique of the concept of scale-
This meta-analysis includes several studies which analyse the neutrality as applied to contemporary African contexts. Fischer
effect of land size and control for potential endogeneity arising argues that the widespread adoption of crop technologies by small-
from unobserved time-invariant variables by applying Mundlak- holders during the Asian green revolution was a result of the ‘‘so-
Chamberlain fixed effects (Ogada et al., 2014; Mukasa, 2018; cial and biological shaping” of crop technology rather than the
Khonje, Manda, Alene, & Kassie, 2015; Wossen et al., 2015; inherent scale-neutrality of the technologies. Contextual differ-
Arslan, McCarthy, Lipper, Asfaw, & Cattaneo, 2014), and one study ences between African agricultural systems today and Asian agri-
which uses Mundlak correlated random effects (Arslan et al., culture during the green revolution, including reduced labour
2017). The technologies investigated in these studies include availability, reduced public agricultural funding, and the privatiza-
NRM, improved varieties, and chemical inputs. With the exception tion of agricultural research and associated restrictions on seed
of Khonje et al. (2018), these authors’ results are consistent with exchange, will cause the benefits of crop technologies to accrue
the mean results for land size found in the present study. primarily to wealthier farmers with larger landholdings.
9
Fig. 3. Bootstrapped expected values of PCC for predictors in the categories of Financial Characteristics and External Factors. The expected value of the effect of each predictor
on adoption of Natural Resource Management (green), Improved Variety (orange), Chemical Inputs (purple) and Mechanization and Infrastructure (magenta) is shown as a
solid square with error bars indicating 95% confidence intervals. The numbers inside the squares indicate the number of studies used in the meta-regression. Non-significant
results are dotted. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
On the other hand, the negative (though not significant) effect Feder (1982) argues that in the simplest case of a single divisi-
of land size on NRM technologies suggests that these technologies ble technology, lack of access to credit should not affect adoption.
are scale-neutral, or perhaps even more readily adopted by small- Even households with very little cash will invest in technologies
holders. Knowler and Bradshaw (2007), in their meta-analysis, find that provide high marginal returns. However, he shows that in
that land size positively influences adoption of conservation agri- the case of two complementary innovations (eg. improved varieties
culture in North America (consistent with the results of and tube wells) a lack of credit inhibits adoption of the lumpy (in-
Baumgart-Getz et al. (2012) in the US and Wauters & Mathijs divisible) investment (the tube well) which lowers the expected
(2014) in the developed world), but not in the developing world, returns from (and discourages adoption of) the divisible technol-
(consistent with the findings of Pattanayak et al., 2003). The dis- ogy. This is supported by the earlier empirical work of Clay (1975).
crepancy may be due to the prevalence of mechanization in the The variable definition for access to credit has received some
developed world, which can encourage adoption of NRM tech- attention in recent empirical studies. While a majority of the liter-
niques at scale. In the developing world, where much agricultural ature uses a dummy variable that takes a value of 1 if a household
labour is done by hand, the labour cost per hectare of establishing has access to credit, Simtowe, Zeller, and Diagne (2009) point out
NRM techniques may be scale-neutral. On the other hand, it is pos- that this will only affect adoption if the household is credit con-
sible that a high prevalence of contract work and labour-sharing strained. They identify households that had an unmet credit need,
agreements in some developing countries allows farmers with and use a treatment effects model to show that credit constraints
small farms to adopt NRM practices by asking or paying a neigh- significantly inhibit adoption of hybrid maize in Malawi. Further-
bour with the necessary machinery. more, they show that credit constraints are highly endogenous,
Access to credit is a variable that has received increased empir- and failing to correct for this endogeneity produces biased results.
ical attention in recent years. Along with access to extension, it is A similar credit constraint definition is used by Teklewold et al.
one of the most likely variables to be influenced by policy. It is pos- (2013), Kassie et al. (2015), Kabunga et al. (2012), Matuschke and
itive in all cases, and significant for all categories except mecha- Qaim (2009), Cavatassi, Lipper, and Narloch (2011), Fisher et al.
nization and infrastructure. The positive effect of credit on (2015), Zeng, Norton, Jaleta, Shiferaw, and Yirga (2018).
technology adoption is consistent with earlier theoretical and Table 5 shows the results of 4 meta-regression models which
empirical work. compare studies that use the credit constraint framework to stud-
10
Table 5
Models of the effect of credit access on adoption of improved varieties. Studies that use the credit constraint framework are found to result in highly significant positive effects,
while other variable definitions are not. This finding is robust to the inclusion of regional effects (Model 2).
Model 0 Model 1 Model 1 Model 2

(no intercept)
Intercept 0.047** 0.028 0.014
(0.015) (0.020) (0.029)
Credit (access framework) 0.028
(0.020)
Credit (constraint framework) 0.088** 0.060* 0.095*
(0.029) (0.035) (0.041)
Credit (other) 0.044 0.016 0.004
(0.034) (0.039) (0.042)
Regional effects1 No No No Yes
N 39 39 39 39
I2 0.87 0.86 0.86 0.81
R2 0.03 0.26
s2 0.006 0.006 0.006 0.005
(0.002) (0.002) (0.002) (0.002)
QE (residual heterogeneity) 204 192 192 117
(df = 38, p <.0001) (df = 36, p <.0001) (df = 36, p < :0001) (df = 29, p <.0001)
QM (moderators) 13.0 2.9 17.7
(df = 3, p ¼ :005) (df = 2, p ¼ :23) (df = 9, p ¼ :04)

p < .001; p < .01; p < .05; y p<.1; Standard errors shown in brackets.
1
This is a factor variable that takes a value based on the regions listed in Table 4. The coefficient estimates for each region are not generalizable beyond the countries
included in the sample and are therefore omitted from the table.
ies that do not. Model 1 (no intercept) shows that the credit con- erosion control, and fertility management technologies. Although
straint definition leads to significant and positive estimates for land tenure is found to be significant for the adoption of NRM
the effect of a (lack of a) credit constraint on adoption of improved technologies as a whole, controlling for the subcategory shows that
varieties (PCC ¼ 0:088 0:057), while the traditional ‘‘access to land tenure is significant for erosion control (PCC ¼ 0:054 0:022)
credit” definition leads to a lower and non-significant estimate and fertility management (PCC ¼ 0:027 0:017), but not for
(PCC ¼ 0:028 0:039). Model 1, where the access to credit frame- conservation tillage (PCC ¼ 0:007 0:017). The result for erosion
work is specified as the intercept, shows that the credit constraint control is relatively robust to the inclusion of controls for
definition leads to larger coefficient estimates (p<.1). Model 2 geographic study area and region, while the result for fertility
shows that this finding is robust to the inclusion of regional effects. management is not.
Ashraf, Giné, and Karlan (2009) provide an illustration of how Land tenure can extend farmers’ planning horizons, which
credit constraints are categorically different from access to credit. affect NRM practices differently from other technologies. Many
Based on a randomized control trial in Kenya, they find that the natural resource management technologies, particularly erosion
provision of credit increased participation in an export market control techniques, require large upfront investments in labour
access program but was not related to increased incomes. They or capital, and often do not yield immediate productivity benefits.
conclude that lack of access to credit may not have been the pri- Lack of secure tenure may thus make the investment prohibitively
mary reason farmers were not accessing export markets. Within risky. For example, Gebremedhin and Swinton (2003) find that
the adoption literature, the consideration of binding constraints, land-secure farmers in the Tigray region of Ethiopia are more likely
or prerequisites, to adoption (rather than simply determinants) to invest in costly stone terraces, while farmers with only short-
has received some attention (e.g Moser & Barrett (2006), term land security are more likely to invest in cheaper but less dur-
Shiferaw et al. (2008), Shiferaw et al. (2015)) which is positive able soil bunds.
trend that may produce results that are more immediately useful Access to extension and organization membership are posi-
to policy makers. tive and significant across four and three out of four technology
Land tenure is found to be significant only in the adoption of categories, respectively. However, causal interpretations of these
NRM technologies. The non-significant effects for improved seeds, variables are in general not appropriate, due to a high risk of endo-
chemical inputs, and mechanization and infrastructure are consis- geneity. Extension workers may seek out farmers they believe will
tent with the lack of agreement in the theoretical literature dis- be likely adopters, or likely adopters may seek out extension ser-
cussed by Feder et al. (1985). Feder and Umali (1993) find that vices. Membership in farmers’ organizations is often considered a
land tenure was a significant determinant of adoption in the early form of social capital, but these organizations are also often used
stages of green revolution technology diffusion, but its importance to disseminate information about new technologies, so this vari-
diminished once the technologies had been widely adopted. able may in some cases be a proxy for awareness of an innovation.
The positive effect of land tenure on NRM technology adoption Even in cases of total population awareness of a technology, self-
agrees with most previous meta-analyses. Pattanayak et al. (2003) selection into farmers’ organizations makes causal inference inap-
find an unambiguous pattern of positive and significant effects on propriate. Di Falco, Feri, Pin, and Vollenweider (2018) even find
adoption of agroforestry. Baumgart-Getz et al. (2012) find a posi- evidence that could imply a reverse (and negative) causation
tive and significant average effect, but acknowledge that there is between social capital and adoption: in a randomized control trial,
a large amount of heterogeneity in the data. Knowler and farmers who received improved varieties of maize reduced their
Bradshaw (2007) find ambiguous results. The variability in previ- interaction with their social network. However, several authors
ous results may be related to the diversity of technologies that have addressed the potential endogeneity of these two predictors.
comprise this category. Studies that explicitly account for sources of endogeneity asso-
Table 6 shows five meta-regression models that are used to ciated with access to extension and organization membership have
divide NRM technologies into subcategories of conservation tillage, found results that are consistent with the average results pre-
11
Table 6
Models of the effect of land tenure on adoption of NRM technologies. Land tenure is found to have a significant effect on adoption of erosion control and fertility management
technologies, but not conservation tillage technologies (model 1 - no intercept). However, only the comparison of erosion control with conservation tillage technologies is found
to be significant (model 1). Studies that cover a large geographic range are found to result in significantly lower estimates.
Model 0 Model 1 Model 1 Model 2 Model 3

(no intercept)
Intercept 0.046*** 0.007 0.030y 0.049
(0.013) (0.017) (0.018) (0.034)
Tech: Cons. Tillage 0.007
(0.017)
Tech: Erosion Control 0.054*** 0.047* 0.042* 0.038y
(0.012) (0.021) (0.019) (0.021)
Tech: Fert. Mgmt. 0.027** 0.020 0.018 0.017
(0.009) (0.020) (0.017) (0.019)
Study Size: > Regional2 0.033* 0.050
(0.013) (0.030)
Regional effects No No No No Yes
N 37 35 35 35 35
I2 0.87 0.53 0.53 0.42 0.52
R2 0.21 0.48 0.31
s2 0.0043 0.0007 0.0007 0.0005 0.0006
(0.0014) (0.0004) (0.0004) (0.0003) (0.0004)
QE 118.1 69.9 69.9 57.0 55.5
(df = 36, p <.0001) (df = 32, p ¼ :0001) (df = 32, p ¼ :0001) (df = 31, p ¼ :003) (df = 27, p ¼ :001)
QM 30.8 6.2 14.2 13.7
(df = 3, p < :0001) (df = 2, p ¼ :045) (df = 3, p ¼ :003) (df = 7, p ¼ :06)

2
This variable indicates whether the subject geographic area comprises multiple administrative regions/states/provinces. Table 4 shows the distribution of studies based
on geographic study size.
sented here. Wossen et al. (2017) use Propensity Score Matching, agricultural technologies. However, the effects are small on aver-
Inverse Probability Weighted Regression Adjustment, and Endoge- age and highly variable between studies. Although it is beyond
nous Switching Regression to control for observable and unobserv- the scope of this paper to determine whether the small average
able factors that affect farmers’ access to extension and effect of extension services on technology adoption justifies their
membership in farmers’ cooperatives. Their results demonstrate, expense, the large amount of variability shows that some services
to p < :01, causal links between adoption of improved varieties are far more effective than others. Future work should focus on
of cassava and both extension access and cooperative membership understanding how methodological differences in extension ser-
in Nigeria. Walisinghe, Ratnasiri, Rohde, and Guest (2017) find that vices relate to their effectiveness in encouraging adoption.
using an instrumental variable to control for endogeneity increases Extension is often considered a substitute for formal education
the estimated marginal effect of extension access for eight different in encouraging adoption (Feder et al., 1985; Nkamleu & Adesina,
technologies in Sri Lanka. However, their choice of instrument (re- 2000). In order to test this hypothesis, the PCC of extension for
gional average extension access) may be inappropriate if there is NRM and IV technologies was regressed on the Education Index.
significant inter-regional knowledge spill-over from farmers with Results are shown in Table 7, NRM Model 1 and IV Model 1. Addi-
access to extension to those without. Wossen et al. (2015) find that tionally, it was hypothesized that the effect of extension would be
instrumenting for membership in informal labour-sharing arrange- reduced for cases where the population adoption level is high, and
ments actually increases the value of the estimated coefficient. technologies may be diffusing outside of formal extension chan-
It may be reasonable to cautiously conclude that extension nels. The population adoption level is included in NRM Model 2
access and organization membership do encourage adoption of and IV Model 2. Interestingly, the correlation with Education Index
Table 7
Models of the effect of access to extension on adoption. Extension is found to have a significant effect on adoption of improved varieties in regions where education levels are low,
but the effect disappears for higher education indices. The opposite is true of NRM technologies, where extension is found to be more effective in regions with higher levels of
education. The Education Index is scaled to zero mean and unit variance.
NRM Model 1 NRM Model 2 IV Model 1 IV Model 2

Intercept 0.034 -0.052 0.244*** 0.219***
(0.040) (0.034) (0.052) (0.060)
Education Index 0.185* 0.194** 0.390** 0.381**
(0.094) (0.075) (0.123) (0.130)
Adoption Level 0.016 0.036
(0.048) (0.064)
N 38 34 61 55
I2 0.84 0.70 0.83 0.85
R2 0.13 0.27 0.19 0.15
s2 0.0037 0.0016 0.007 0.008
(0.0013) (0.0007) (0.002) (0.002)
QE 109.8 81.4 230.8 220.2
(df = 36, p < :0001) (df = 31, p < :0001) (df = 59, p < :0001) (df = 52, p < :0001)
QM 3.9 7.1 10.0 8.7
(df = 1, p ¼ :049) (df = 2, p ¼ :029) (df = 1, p ¼ :002) (df = 2, p ¼ :013)

12
is positive for NRM technologies, and negative for IV technologies. referred to the following literature, which is primarily focused on
This indicates that the effect of extension on adoption of improved Sub-Saharan Africa: Atwood (1990), Pinckney and Kimuyu
varieties is greatest for areas with low levels of education, as may (1994), Firmin-sellers and Sellers (1999), Toulmin (2008), Sitko,
be expected if the two factors can substitute for each other. How- Chamberlin, and Hichaambwa (2014), Asaaga, Hirons, and Malhi
ever, for NRM technologies, it appears that the effectiveness of (2020).
extension is augmented when education levels are higher. A possi- Credit is found to encourage the adoption of all technologies,
ble explanation is that many NRM technologies are more compli- although the effect for mechanization and infrastructure is non-
cated to adopt. Farmers may be more likely to adopt significant. In addition, studies that use a credit constraint frame-
technologies when they feel they understand them, and farmers work generally find that credit plays a larger role in adoption than
with primary or secondary education may be better able to under- studies that use the simpler access to credit definition. Credit
stand the information provided by extension agents. The effect of schemes that target farmers with unmet credit needs can be ben-
extension was not found to vary significantly with the population eficial to the dissemination of agricultural technologies. Further,
adoption level. adoption researchers should begin to use the credit constraint
framework in place of the access to credit framework, where data
permit.
5. Conclusions and recommendations Given the biases that are introduced by model specifications
and variable definitions, it is recommended that researchers
This study has presented a quantitative meta-analysis of the explicitly identify the causal pathways they intend to test. These
effect size of 15 predictor variables on adoption of 4 categories of pathways may be based on classical adoption theories (such as
technology using the largest dataset to date. The results are mostly those presented by Feder et al. (1985) and Rogers (2003)), as well
consistent with previous semi-quantitative meta-analyses, and as intimate knowledge of the local context. Researchers should also
agree that there are few universal trends to discover. While signif- test a number of models, identify the assumptions made in each,
icant effects are found in the average results, most of these aver- attempt to link these assumptions to any differences in the results,
ages are dominated by heterogeneity. However, the average and admit the uncertainty introduced by any ambiguity in the
results that are (relatively) large and significant are highly consis- results. Ultimately, a thorough exploration of the data structure
tent with adoption theory. Education, land size, access to credit, (through covariance matrices, etc.) may be more useful than the
land tenure, contact with extension agents, and membership in results of a single correlational model.
farmers’ organizations all positively influence the adoption of most Determinants of adoption are highly dependent on unobserved
of the technology categories considered here. Based on the key cultural, contextual, and policy factors, which is evidenced by the
results, insights and recommendations can be shared. small average effects, the large amount of unexplained heterogene-
Education positively and significantly influences the adoption of ity in all of the average results presented, and the inability of
improved varieties, chemical inputs, and mechanization and observed factors to explain much of this variability. Policy makers
infrastructure. These results highlight the potential for education should combine the high-level results presented here with site-
to improve farmer capacity to ‘‘respond to new events in the con- specific information about the places where they hope to promote
text of risk” (Schultz, 1981). Extension is found to substitute for adoption. Ultimately, there can be no substitute for an intimate
education in the case of improved varieties. Education is not found understanding of, and relationship with, the people and culture
to affect adoption of NRM technologies on average. However, of a region.
extension is found to be most effective in areas with high levels
of education, which suggests that the two variables are comple- CRediT authorship contribution statement
mentary. Alternatively, this suggests that extension strategies for
NRM technologies have been ineffective in regions with low levels Sacha Ruzzante: Conceptualization, Methodology, Formal anal-
of education. Considering the continued low levels of adoption ysis, Data curation, Writing - original draft. Ricardo Labarta:
(Grabowski et al., 2016), innovations in information delivery meth- Methodology, Resources, Supervision, Writing - review & editing.
ods may be required. Continued investment in education in the Amy Bilton: Conceptualization, Methodology, Resources, Writing
developing world is crucial for many reasons, only one of which - review & editing, Supervision, Funding acquisition.
is increasing the ability of farmers to understand and make
informed decisions about new technologies. Declaration of Competing Interest
The results for land size suggest that improved varieties may
not be scale-neutral, as is generally believed. Despite international The authors declare that they have no known competing finan-
recognition of the need to develop technologies appropriate for cial interests or personal relationships that could have appeared
smallholders, larger farmers are still more likely to use the prod- to influence the work reported in this paper.
ucts of modern agricultural research. More attention should be
focused on eliminating fixed adoption costs for smallholders. Acknowledgements
Land tenure is found to be more important for the adoption of
erosion control technologies, which are longer-term investments, We wish to thank Dr. Naoya Abe, for the helpful discussions
than for other NRM technologies. Where erosion of agricultural regarding the methods in the study and Dr. David Taylor, whose
land is a significant issue, development practitioners should assess comments and suggestions helped improve and clarify the manu-
whether lack of tenure security is limiting the adoption of benefi- script. We also wish to thank the authors of primary studies who
cial erosion control techniques. Efforts to grant land titles, or to responded to requests for clarification of results published in their
otherwise provide tenure security to tenure-insecure farmers studies.
may yield societal and environmental benefits that range far This research was supported in part by the following:
beyond the individual. It is noted that there is a large and well- Powering Agriculture: An Energy Grand Challenge for Devel-
developed debate about appropriate strategies for securing tenure opment. Partners: the United States Agency for International
rights for farmers, and indeed whether formal land titles are Development (USAID), the Swedish Government, the German Fed-
always required to ensure tenure security. A discussion of this eral Ministry for Economic Cooperation and Development (BMZ),
debate is beyond the scope of the current paper, but the reader is
13
Duke Energy, and the Overseas Private Investment Corporation Cavatassi, R., Lipper, L., & Narloch, U. (2011). Modern variety adoption and risk
management in drought prone areas: Insights from the sorghum farmers of
(OPIC) under grant number AID-OAA-A-16-00005.
eastern Ethiopia. Agricultural Economics, 42(3), 279–292. https://doi.org/
University of Toronto Centre for Global Engineering, Metcalfe 10.1111/j.1574-0862.2010.00514.x.
Family Graduate Scholarship. Clay, E. J. (1975). Equity and productivity effects of a package of technical
Natural Sciences and Engineering Research Council, Canada innovations and changes in social institutions: Tubewells, tractors, and high-
yielding varieties. Indian Journal of Agricultural Economics, XXX(1), 74–87.
Graduate Scholarship. Cohen, J. (1988). Statistical Power Analysis for Behavioral Sciences. New York:
Government of Ontario, Canada, Ontario Graduate Lawrence Erlbaum Associates, Publishers, 2nd ed..
Scholarship. Dadi, L., Burton, M., & Ozanne, A. (2004). Duration analysis of technological
adoption in Ethiopian agriculture. Journal of Agricultural Economics, 55(3),
613–631. https://doi.org/10.1111/j.1477-9552.2004.tb00117.x.
Appendix A. Supplementary data Dalton, T. J., Lilja, N. K., Johnson, N., & Howeler, R. (2011). Farmer participatory
research and soil conservation in Southeast Asian Cassava Systems. World
Development, 39(12), 2176–2186. https://doi.org/10.1016/
Supplementary data associated with this article can be found, in j.worlddev.2011.05.011.
the online version, athttps://doi.org/10.1016/j.worlddev.2021. DerSimonian, R., & Kacker, R. (2007). Random-effects model for meta-analysis of
clinical trials: An update. Contemporary Clinical Trials, 28(2), 105–114. https://
105599.
doi.org/10.1016/j.cct.2006.04.004.
Di Falco, S., & Bulte, E. (2013). The impact of kinship networks on the adoption of
References risk-mitigating strategies in Ethiopia. World Development, 43, 100–110. https://
doi.org/10.1016/j.worlddev.2012.10.011.
Di Falco, S., Feri, F., Pin, P., & Vollenweider, X. (2018). Ties that bind: Network
Adeoti, A. I. (2008). Factors influencing irrigation technology adoption and its
redistributive pressure and economic decisions in village economies. Journal of
impact on household poverty in Ghana. Journal of Agriculture and Rural
Development Economics 131(November 2017), 123–131.https://doi.org/10.
Development in the Tropics and Subtropics, 109(1), 51–63.
1016/j.jdeveco.2017.12.001..
Adesina, A. A. (1996). Factors affecting the adoption of fertilizers by rice farmers in
Diagne, A., & Demont, M. (2007). Taking a new look at empirical models of
Cote d’Ivoire. Nutrient Cycling in Agroecosystems, 46, 29–39.
adoption: Average treatment effect estimation of adoption rates and their
Adesina, A. A., & Zinnah, M. (1993). Technology characteristics, farmers’ perceptions
determinants. Agricultural Economics, 37(2–3), 201–210. https://doi.org/
and adoption decisions: A Tobit model application in Sierra Leone. Agricultural
10.1111/j.1574-0862.2007.00266.x.
Economics, 9(4), 297–311.
Doss, C. R. (2006). Analyzing technology adoption using microstudies: Limitations,
Alabi, O. O., Lawal, A. F., Coker, A. A., & Awoyinka, Y. A. (2014). Probit model analysis
challenges, and opportunities for improvement. Agricultural Economics, 34(3),
of smallholder’s farmers decision to use agrochemical inputs in Gwagwalada
207–219. https://doi.org/10.1111/j.1574-0864.2006.00119.x.
and Kuje Area Councils of Federal Capital Territory, Abuja, Nigeria. International
Doucouliagos, H. C. (2011). How Large is Large? Preliminary and relative guidelines
Journal of Food and Agricultural Economics, 2(1), 85–93. https://doi.org/10.1073/
for interpreting partial correlations in economics..
pnas.1011013108.
Feder, G. (1982). Adoption of interrelated agricultural innovations:
Ali, A., Hussain, I., Rahut, D. B., & Erenstein, O. (2018). Laser-land leveling adoption
Complementarity and the impacts of risk, scale, and credit. American Journal
and its impact on water use, crop yields and household income: Empirical
of Agricultural Economics, 64(1), 94–101.
evidence from the rice-wheat system of Pakistan Punjab. Food Policy, 77, 19–32.
Feder, G., Just, R. E., & Zilberman, D. (1985). Adoption of agricultural innovations in
https://doi.org/10.1016/j.foodpol.2018.03.018.
developing countries: A survey. Economic Development and Cultural Change, 33
Amsalu, A., & de Graaff, J. (2007). Determinants of adoption and continued use of
(2), 255–298.
stone terraces for soil and water conservation in an Ethiopian highland
Feder, G., & Umali, D. L. (1993). The adoption of agricultural innovations: A review.
watershed. Ecological Economics, 61(2–3), 294–302. https://doi.org/10.1016/j.
Technological Forecasting and Social Change, 43, 215–239.
ecolecon.2006.01.014.
Firmin-sellers, K., & Sellers, P. (1999). Expected failures and unexpected successes of
Andersson, J. A., & D’Souza, S. (2014). From adoption claims to understanding
land titling in Africa. World Development, 27(7), 1115–1128. https://doi.org/
farmers and contexts: A literature review of Conservation Agriculture (CA)
10.1016/S0305-750X(99)00058-3.
adoption among smallholder farmers in southern Africa. Agriculture, Ecosystems
Fischer, K. (2016). Why new crop technology is not scale-neutral – A critique of the
and Environment, 187, 116–132. https://doi.org/10.1016/j.agee.2013.08.008.
expectations for a crop-based African Green Revolution. Research Policy, 45(6),
Arslan, A., Belotti, F., & Lipper, L. (2017). Smallholder productivity and weather
1185–1194. https://doi.org/10.1016/j.respol.2016.03.007.
shocks: Adoption and impact of widely promoted agricultural practices in
Fisher, M., Abate, T., Alemayehu, Y., & Madulu, R. B. (2015). Drought tolerant maize
Tanzania. Food Policy, 69, 68–81. https://doi.org/10.1016/j.foodpol.2017.03.005.
for farmer adaptation to drought in sub-Saharan Africa: Determinants of
Arslan, A., McCarthy, N., Lipper, L., Asfaw, S., & Cattaneo, A. (2014). Adoption and
adoption in eastern and southern Africa. Climatic Change, 133(2), 283–299.
intensity of adoption of conservation farming practices in Zambia. Agriculture,
https://doi.org/10.1007/s10584-015-1459-2.
Ecosystems and Environment, 187, 72–86. https://doi.org/10.1016/j.
Floro, V. O., Labarta, R. A., Becerra López-Lavalle, L. A., Martinez, J. M., & Ovalle, T. M.
agee.2013.08.017.
(2018). Household determinants of the adoption of improved cassava varieties
Asaaga, F. A., Hirons, M. A., & Malhi, Y. (2020). Questioning the link between tenure
using DNA fingerprinting to identify varieties in farmer fields: A case study in
security and sustainable land management in cocoa landscapes in Ghana. World
Colombia. Journal of Agricultural Economics, 69(2), 518–536. https://doi.org/
Development, 130. https://doi.org/10.1016/j.worlddev.2020.104913 104913.
10.1111/1477-9552.12247.
Ashraf, N., Giné, X., & Karlan, D. (2009). Finding missing markets (and a disturbing
Friedman, L. (2001). Why Vote Count Reviews Don’t Count. Biological Psychiatry 49
epilogue): Evidence from an export crop adoption and marketing intervention
(January 2001) 161–163.https://doi.org/10.1016/S0006-3223(00)01075-1..
in Kenya. American Journal of Agricultural Economics, 91(4), 973–990. https://doi.
Gebregziabher, G., Giordano, M. A., Langan, J. S., & Namara, R. E. (2014). Economic
org/10.1111/j.l467-8276.2009.01319.x.
analysis of factors influencing adoption of motor pumps in Ethiopia. Journal of
Atwood, D. A. (1990). Land registration in Africa: The impact on agricultural
Development and Agricultural Economics, 6(90), 490–500.
production. World Development, 18(5), 659–671. https://doi.org/10.1016/0305-
Gebremedhin, B., & Swinton, S. M. (2003). Investment in soil conservation in
750X(90)90016-Q.
northern Ethiopia: The role of land tenure security and public programs.
Audu, V. I., & Aye, G. C. (2014). The effects of improved maize technology on
Agricultural Economics, 29, 69–84. https://doi.org/10.1016/S0169-5150(03)
household welfare in Buruku, Benue State, Nigeria. Cogent Economics and
00022-7.
Finance, 2(1), 1–10. https://doi.org/10.1080/23322039.2014.960592.
Grabowski, P. P., Kerr, J. M., Haggblade, S., & Kabwe, S. (2016). Determinants of
Baumgart-Getz, A., Stalker, L., & Floress, K. (2012). Why farmers adopt best
adoption and disadoption of minimum tillage by cotton farmers in eastern
management practice in the United States: A meta-analysis of the adoption
Zambia. Agriculture, Ecosystems and Environment, 231, 54–67. https://doi.org/
literature. Journal of Environmental Management, 96(1), 17–25. https://doi.org/
10.1016/j.agee.2016.06.027.
10.1016/j.jenvman.2011.10.006.
Hedges, L. V., & Olkin, I. (1980). Vote-counting methods in research synthesis.
Baumgartner, P., & Cherlet, J. (2016). Institutional Framework of (In)Action Against
Psychological Bulletin, 88(2), 359–369.
Land Degradation. In E. Nkonya, A. Mirzabaev, & J. von Braun (Eds.), Economics of
Higgins, J. P. T., & Thompson, S. G. (2002). Quantifying heterogeneity in a meta-
Land Degradation and Improvement – A Global Assessment for Sustainable
analysis. Statistics in Medicine, 21, 1539–1558. https://doi.org/10.1002/
Development (pp. 33–54). Cham: Springer.
sim.1186.
Bokusheva, R., Finger, R., Fischler, M., Berlin, R., Marín, Y., Pérez, F., & Paiz, F. (2012).
Kabunga, N. S., Dubois, T., & Qaim, M. (2012). Heterogeneous information exposure
Factors determining the adoption and impact of a postharvest storage
and technology adoption: The case of tissue culture bananas in Kenya.
technology. Food Security, 4(2), 279–293. https://doi.org/10.1007/s12571-012-
Agricultural Economics (United Kingdom), 43(5), 473–486. https://doi.org/
0184-1.
10.1111/j.1574-0862.2012.00597.x.
Burton, R. J. F. (2014). The influence of farmer demographic characteristics on
Kassie, M., Teklewold, H., Jaleta, M., Marenya, P., & Erenstein, O. (2015).
environmental behaviour: A review. Journal of Environmental Management, 135,
Understanding the adoption of a portfolio of sustainable intensification
19–26. https://doi.org/10.1016/j.jenvman.2013.12.005.
practices in eastern and southern Africa. Land Use Policy, 42, 400–411. https://
Bushman, B. J., & Wang, M. C. (1995). A procedure for combining sample correlation
doi.org/10.1016/j.landusepol.2014.08.016.
coefficients and vote counts to obtain an estimate and a confidence interval for
Katengeza, S., Kankwamba, H., & Mangisoni, J. H. (2015). Adoption Pathways project
the population correlation coefficient. Quantitative Methods in Psychology, 117
discussion paper 7..
(3), 530–546. https://doi.org/10.1037/0033-2909.117.3.530.
14
Katengeza, S. P., Holden, S. T., & Lunduka, R. W. (2018). Adoption of drought tolerant Ogada, M. J., Mwabu, G., & Muchai, D. (2014). Farm technology adoption in Kenya: A
maize varieties under rainfall stress in Malawi. Journal of Agricultural Economics, simultaneous estimation of inorganic fertilizer and improved maize variety
70(1), 1–17. https://doi.org/10.1111/1477-9552.12283. adoption decisions. Agricultural and Food Economics, 2(12), 1–28. https://doi.
Khonje, M., Manda, J., Alene, A. D., & Kassie, M. (2015). Analysis of adoption and org/10.1186/s40100-014-0012-3.
impacts of improved maize varieties in Eastern Zambia. World Development, 66 Ogundari, K., & Bolarinwa, O. D. (2018). Impact of agricultural innovation adoption:
(695). https://doi.org/10.1016/j.worlddev.2014.09.008. 706–706. a meta-analysis. Australian Journal of Agricultural and Resource Economics, 62(2),
Khonje, M. G., Manda, J., Mkandawire, P., Tufa, A. H., & Alene, A. D. (2018). Adoption 217–236. https://doi.org/10.1111/1467-8489.12247.
and welfare impacts of multiple agricultural technologies: evidence from Ojiako, I. A., Manyong, V. M., & Ikpi, A. E. (2007). Determinants of rural farmers’
eastern Zambia. Agricultural Economics, 49, 599–609. https://doi.org/10.1111/ improved soybean adoption decisions in northern Nigeria. Journal of Food,
agec.12445. Agriculture and Environment, 5(2), 215–223.
Kivlin, J. E., & Fliegel, F. C. (1967). Differential perceptions of innovations and rate of Orr, A., Kambombo, B., Roth, C., Harris, D., & Doyle, V. (2015). Adoption of Integrated
adoption. Rural Sociology, 32(1), 78–91. Food-Energy Systems: Improved Cookstoves and Pigeonpea in Southern
Knight, J., Weir, S., & Woldehanna, T. (2003). The role of education in facilitating Malawi. Experimental Agriculture, 51(2), 191–209. https://doi.org/10.1017/
risk-taking and innovation in agriculture. Journal of Development Studies, 39(6), S0014479714000222.
1–22. https://doi.org/10.1080/00220380312331293567. Pardey, P. G., Andrade, R. S., Hurley, T. M., Rao, X., & Liebenberg, F. G. (2016). Returns
Knowler, D., & Bradshaw, B. (2007). Farmers’ adoption of conservation agriculture: to food and agricultural R & D investments in Sub-Saharan Africa, 1975–2014.
A review and synthesis of recent research. Food Policy, 32(1), 25–48. https://doi. Food Policy, 65, 1–8. https://doi.org/10.1016/j.foodpol.2016.09.009.
org/10.1016/j.foodpol.2006.01.003. Pardey, P. G., Chan-Kang, C., Dehmer, S. P., & Beddow, J. M. (2016). Agricultural R&D
Kristjanson, P., Okike, I., Tarawali, S., Singh, B. B., & Manyong, V. M. (2005). Farmers ’ is on the move. Nature, 537, 301–303.
perceptions of benefits and factors affecting the adoption of improved dual- Pattanayak, S. K., Mercer, D. E., Sills, E., & Yang, J. C. (2003). Taking stock of
purpose cowpea in the dry savannas of Nigeria. Agricultural Economics, 32, agroforestry adoption studies. Agroforestry Systems, 57(3), 173–186. https://doi.
195–210. org/10.1023/A:1024809108210.
Lambrecht, I., Vanlauwe, B., Merckx, R., & Maertens, M. (2014). Understanding the Paule, R. C., & Mandel, J. (1982). Consensus values and weighting factors. Journal of
process of agricultural technology adoption: Mineral fertilizer in Eastern DR Research of the National Bureau of Standards, 87(5), 377–385. https://doi.org/
Congo. World Development, 59, 132–146. https://doi.org/10.1016/ 10.6028/jres.087.022.
j.worlddev.2014.01.024. Pinckney, T. C., & Kimuyu, P. K. (1994). Land tenure reform in East Africa: Good, bad,
Larsen, A. F. (2018). When knowledgable neighbors also share seedlings: Diffusion or unimportant? Journal of African Economies, 3(1), 1–28. https://doi.org/
of banana cultivation in tanzania. Agricultural Economics (ja), 1–15. https://doi. 10.1093/oxfordjournals.jae.a036794.
org/10.1111/agec.12465. Prager, K., & Posthumus, H. (2010). Socio-economic factors influencing farmers’
Liu, T., Bruins, R. J. G., & Heberling, M. T. (2018). Factors influencing farmers’ adoption of soil conservation practices in Europe. In T. L. Napier (Ed.), Human
adoption of best management practices: A review and synthesis. Sustainability, dimensions of soil and water conservation, number January chapter 12
10(2), 1–26. https://doi.org/10.3390/su10020432. (pp. 203–223). New York: Nova Science Publishers Inc.
Lowder, S. K., Skoet, J., & Raney, T. (2016). The number, size, and distribution of Ranganathan, P., Pramesh, C. S., & Buyse, M. (2015). Common pitfalls in statistical
farms, smallholder farms, and family farms worldwide. World Development, 87, analysis: ’No evidence of effect’ versus ‘evidence of no effect’, 6(1), 62–63. https://
16–29. https://doi.org/10.1016/j.worlddev.2015.10.041. doi.org/10.4103/2229-3485.148821.
Maredia, M. K., & Anthony, D. (2010). Estimating overall returns to international Rapsomanikis, G. (2015). The economic lives of smallholder farmers (Technical
agricultural research in Africa through benefit-cost analysis: A ‘best-evidence’ report). Rome: Food and Agriculture Organization of the United Nations.
approach. Agricultural Economics, 41, 81–100. https://doi.org/10.1111/j.1574- Rogers, E. M. (2003). Diffusion of Innovations (5th ed.). New York: Simon & Schuster.
0862.2009.00427.x. Rubas, D. (2004). Technology adoption: Who is likely to adopt and how does the
Marenya, P. P., & Barrett, C. B. (2007). Household-level determinants of adoption of timing affect the benefits (Ph.D. thesis). Texas A&M..
improved natural resources management practices among smallholder farmers Ruzzante, Sacha, & Bilton, Amy (2021). Adoption of Agricultural Technologies in the
in western Kenya. Food Policy, 32, 515–536. https://doi.org/10.1016/ Developing World: A Dataset of Primary Study Results. Data in Brief. In press.
j.foodpol.2006.10.002. Ryan, B., & Gross, N. (1943). The diffusion of hybrid seed corn in two Iowa
Matuschke, I., & Qaim, M. (2009). The impact of social networks on hybrid seed communities. Rural Sociology, 8(1), 15–24.
adoption in India. Agricultural Economics, 40(5), 493–505. https://doi.org/ Schultz, T. W. (1981). Investing in people: The economics of population quality.
10.1111/j.1574-0862.2009.00393.x. Berkeley, CA: University of California Press Ltd.
Mauceri, M., Alwang, J., Norton, G., & Barrera, V. (2007). Effectiveness of integrated Shiferaw, B., Kebede, T., Kassie, M., & Fisher, M. (2015). Market imperfections, access
pest management dissemination techniques. Journal of Applied and Agricultural to information and technology adoption in Uganda: Challenges of overcoming
Economics, 39(3), 765–780. multiple constraints. Agricultural Economics, 46(4), 475–488. https://doi.org/
Meijer, S. S., Catacutan, D., Ajayi, O. C., Sileshi, G. W., & Nieuwenhuis, M. (2015). The 10.1111/agec.12175.
role of knowledge, attitudes and perceptions in the uptake of agricultural and Shiferaw, B. A., Kebede, T. A., & You, L. (2008). Technology adoption under seed
agroforestry innovations among smallholder farmers in sub-Saharan Africa. access constraints and the economic impacts of improved pigeonpea varieties
International Journal of Agricultural Sustainability, 13(1), 40–54. https://doi.org/ in Tanzania. Agricultural Economics, 39(3), 309–323. https://doi.org/10.1111/
10.1080/14735903.2014.912493. j.1574-0862.2008.00335.x.
Moser, C. M., & Barrett, C. B. (2006). The complex dynamics of smallholder Simtowe, F., Zeller, M., & Diagne, A. (2009). The impact of credit constraints on the
technology adoption: The case of SRI in Madagascar. Agricultural Economics, 35 adoption of hybrid maize in Malawi. Review of Agricultural and Environmental
(3), 373–388. https://doi.org/10.1111/j.1574-0862.2006.00169.x. Studies, 90(1), 5–22.
Mottaleb, K. A., Krupnik, T. J., & Erenstein, O. (2016). Factors associated with small- Sitko, N. J., Chamberlin, J., & Hichaambwa, M. (2014). Does smallholder land titling
scale agricultural machinery adoption in Bangladesh: Census findings. Journal of facilitate agricultural growth? An analysis of the determinants and effects of
Rural Studies, 46, 155–168. https://doi.org/10.1016/j.jrurstud.2016.06.012. smallholder land titling in Zambia. World Development, 64, 791–802. https://doi.
Mukasa, A. N. (2018). Technology adoption and risk exposure among smallholder org/10.1016/j.worlddev.2014.07.014.
farmers: Panel data evidence from Tanzania and Uganda. World Development, Smits, J., & Permanyer, I. (2019). The subnational human development database.
105, 299–309. https://doi.org/10.1016/j.worlddev.2017.12.006. Scientific Data, 6. https://doi.org/10.1038/sdata.2019.38 190038.
Mundlak, Y. (1978). On the pooling of time series and cross section data. Sodjinou, E., Glin, L., Nicolay, G., Tovignan, S., & Hinvi, J. (2016). Socioeconomic
Econometrica, 46(1), 69–85. determinants of organic cotton adoption in Benin, West Africa. Agricultural and
Ndiritu, S. W., Kassie, M., & Shiferaw, B. (2014). Are there systematic gender Food Economics, 3(12), 1–22. https://doi.org/10.1186/s40100-015-0030-9.
differences in the adoption of sustainable agricultural intensification practices? Stanley, T. D. (2001). Wheat from chaff: Meta-analysis as quantitative literature
Evidence from Kenya. Food Policy, 49, 117–127. https://doi.org/10.1016/ review. The Journal of Economic Perspectives, 15(3), 131–150.
j.foodpol.2014.06.010. Teklewold, H., Kassie, M., & Shiferaw, B. (2013). Adoption of multiple sustainable
Negatu, W., & Parikh, A. (1999). The impact of perception and other factors on the agricultural practices in rural Ethiopia. Journal of Agricultural Economics, 64(3),
adoption of agricultural technology in the Moret and Jiru Woreda (district) of 597–623. https://doi.org/10.1111/1477-9552.12011.
Ethiopia. Agricultural Economics, 21(2), 205–216. https://doi.org/10.1016/ Teshome, A., de Graaff, J., & Kassie, M. (2016). Household-Level Determinants of Soil
S0169-5150(99)00020-1. and Water Conservation Adoption Phases: Evidence from North-Western
Neupane, R. P., Sharma, K. R., & Thapa, G. B. (2002). Adoption of agroforestry in the Ethiopian Highlands. Environmental Management, 57(3), 620–636. https://doi.
hills of Nepal: A logistic regression analysis. Agricultural Systems, 72(3), org/10.1007/s00267-015-0635-5.
177–196. https://doi.org/10.1016/S0308-521X(01)00066-X. Toulmin, C. (2008). Securing land and property rights in sub-Saharan Africa: The
Njuki, J. M., Mapila, M. T., Zingore, S., & Delve, R. (2008). The dynamics of social role of local institutions. Land Use Policy, 26, 10–19. https://doi.org/10.1016/
capital in influencing use of soil management options in the Chinyanja triangle j.landusepol.2008.07.006.
of Southern Africa. Ecology and Society, 13(2). Verkaart, S., Munyua, B. G., Mausch, K., & Michler, J. D. (2017). Welfare impacts of
Nkamleu, G. B., & Adesina, A. A. (2000). Determinants of chemical input use in peri- improved chickpea adoption: A pathway for rural development in Ethiopia?.
urban lowland systems: Bivariate probit analysis in Cameroon. Agricultural Food Policy, 66, 50–61. https://doi.org/10.1016/j.foodpol.2016.11.007.
Systems, 63(2), 111–121. https://doi.org/10.1016/S0308-521X(99)00074-8. Viechtbauer, W. (2010). Conducting meta-analyses in R with the metafor package.
Ntshangase, N. L., Muroyiwa, B., & Sibanda, M. (2018). Farmers’ perceptions and Journal of Statistical Software, 36(3), 1–48.
factors influencing the adoption of no-till conservation agriculture by small- Wainaina, P., Tongruksawattana, S., & Qaim, M. (2016). Tradeoffs and
scale farmers in Zashuke, KwaZulu-Natal province. Sustainability (Switzerland), complementarities in the adoption of improved seeds, fertilizer, and natural
10(2), 1–16. https://doi.org/10.3390/su10020555.
15
resource management technologies in Kenya. Agricultural Economics, 47(3), on technology adoption and household welfare. Journal of Rural Studies, 54,
351–362. https://doi.org/10.1111/agec.12235. 223–233. https://doi.org/10.1016/j.jrurstud.2017.06.022.
Walisinghe, B. R., Ratnasiri, S., Rohde, N., & Guest, R. (2017). Does agricultural Wossen, T., Berger, T., & Di Falco, S. (2015). Social capital, risk preference and
extension promote technology adoption in Sri Lanka. International Journal of adoption of improved farm land management practices in Ethiopia. Agricultural
Social Economics, 44(12), 2173–2186. https://doi.org/10.1108/IJSE-10-2016- Economics (United Kingdom), 46(1), 81–97. https://doi.org/10.1111/agec.12142.
0275. Yigezu, Y. A., Alwang, J., Rahman, M. W., & Mollah, M. B. R. (2019). Is DNA
Wauters, E., & Mathijs, E. (2014). The adoption of farm level soil conservation fingerprinting the gold standard for estimation of adoption and impacts of
practices in developed countries: A meta-analytic review. International Journal improved lentil varieties? Food Policy 83(April 2018), 48–59.https://doi.org/10.
of Agricultural Resources Governance and Ecology, 10(1), 78–102. https://doi.org/ 1016/j.foodpol.2018.11.004..
10.13140/2.1.3770.1766. Yigezu, Y. A., et al. (2018). Enhancing adoption of agricultural technologies requiring
Wollni, M., Lee, D. R., & Thies, J. E. (2010). Conservation agriculture, organic high initial investment among smallholders. Technological Forecasting and
marketing, and collective action in the Honduran hillsides. Agricultural Social Change 134(November 2017), 199–206.https://doi.org/10.1016/j.
Economics, 41(3–4), 373–384. https://doi.org/10.1111/j.1574- techfore.2018.06.006..
0862.2010.00445.x. Zeng, D., Norton, G., Jaleta, M., Shiferaw, B., & Yirga, C. (2018). Land ownership and
Wossen, T., Abdoulaye, T., Alene, A., Haile, M. G., Feleke, S., Olanrewaju, A., & technology adoption revisited: Improved maize varieties in Ethiopia. Land Use
Manyong, V. (2017). Impacts of extension access and cooperative membership Policy, 72(March 2018), 270–279.https://doi.org/10.1016/j.landusepol.2017.12.
047..
16

Adoption of Agricultur

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Adoption of Agricultur

Uploaded by

Copyright:

Available Formats

World Development 146 (2021) 105599

Contents lists available at ScienceDirect

Adoption of agricultural technology in the developing world:

1. Introduction to develop technologies that increase yields, reduce exposure to

Authors Technology Region Methods Sample Notes

All (N = 367) IV (N = 143) NRM (N = 93) Chem (N = 45) Mech (N = 56)

Model 0 Model 1 Model 1 Model 2

Model 0 Model 1 Model 1 Model 2 Model 3

NRM Model 1 NRM Model 2 IV Model 1 IV Model 2

You might also like