Are Rural Youth in Ethiopia Abandoning Agriculture

World Development 106 (2018) 393–406
Contents lists available at ScienceDirect
World Development
journal homepage: www.elsevier.com/locate/worlddev
Big cities, small towns, and poor farmers: Evidence from Ethiopia
Joachim Vandercasteelen a, Seneshaw Temru Beyene a, Bart Minten b,⇑, Johan Swinnen a
a
LICOS – Center for Institutions and Economic Performance, Department of Economics, University of Leuven, Waaistraat 6, Box 3511, B-3000 Leuven, Belgium
b
Ethiopia Strategy Support Program, Ethiopian Development Research Institute, International Food Policy Research Institute, PO Box 5689, Addis Ababa, Ethiopia
a r t i c l e i n f o a b s t r a c t
Article history: Urbanization is happening fast in the developing world and especially so in sub-Saharan Africa where
Accepted 3 March 2018 growth rates of cities are among the highest in the world. While cities and, in particular, secondary towns,
Available online 17 March 2018 where most of the urban population in sub-Saharan Africa resides, affect agricultural practices in their
rural hinterlands, this relationship is not well understood. To fill this gap, we develop a conceptual model
Keywords: to analyze how farmers’ proximity to cities of different sizes affects agricultural prices and intensification
Urbanization of farming. We then test these predictions using large-scale survey data from producers of teff, a major
Cities
staple crop in Ethiopia, relying on unique data on transport costs and road networks and implementing
Secondary towns
Ethiopia
an array of econometric models. We find that agricultural price behavior and intensification is deter-
Sub-Saharan Africa mined by proximity to a city and the type of city. While proximity to cities has a strong positive effect
Agricultural prices on agricultural output prices and on uptake of modern inputs and yields on farms, the effects on prices
Intensification and intensification measures are lower for farmers in the rural hinterlands of secondary towns compared
to primate cities.
Ó 2018 The Authors. 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 In Africa, the share of the population that is urban, at 40 per-

cent, is lower than in Latin America or Asia, but rapid increases
Urbanization rates are quickly increasing in developing coun- in urbanization rates are anticipated over the next decades, result-
tries, with two-thirds of the world population anticipated to be living in a projected African urban population of 55 percent in 2050
ing in cities by 2050 (UN Population Division, 2014). Urbanization (UN Population Division, 2014). However, the process of urbaniza-
is considered an important long-term driver of economic develop- tion in sub-Saharan Africa is argued as substantially different from
ment as it involves the structural transformation of the economy the rest of the world. First, economic growth in African countries
from being rural and agricultural-based towards one that is mod- has been much slower compared to regions that have experienced
ern, urban, and industrial (Henderson & Wang, 2005). Mostly similar changes in urbanization rates in the past decades (Bloom,
through rural-urban migration, employment typically shifts from Canning, & Fink, 2008; Brückner, 2012). Second, while industrial-
agricultural to more remunerative non-farm activities (Gollin, ization and the creation of non-farm job opportunities have been
Parente, & Rogerson, 2002). Moreover, agglomeration in primate the main drivers behind urbanization in Asia, African urbanization
cities (metropolization) can generate localized external economies has occurred without industrialization. Instead, urbanization has
of scale, technological innovations, industrial clustering or knowl- been induced by population pressure, natural resource exploita-
edge accumulation, and additional employment opportunities tion, climate change, conflicts, and political or ethnical tensions
(Bloom, Canning, & Fink, 2008; Henderson, 2010). Recent evidence (Bloom, Canning, & Fink, 2008; Gollin, Jedwab, & Vollrath, 2016;
has confirmed the positive correlation between urbanization rates Henderson, Storeygard, & Deichmann, 2014). As a consequence,
and income per capita (Dorosh & Thurlow, 2014; Henderson, 2010; there is renewed interest in policy research on how urbanization
Ravallion, Chen, & Sangraula, 2007), although establishing causal- determines the structural process of transforming African econo-
ity remains an important challenge when interpreting these empir- mies (Brückner, 2012; Henderson, Storeygard, & Roberts, 2013).
ical findings (Bloom, Canning, & Fink, 2008). It is not only the aggregate rate of urbanization, but also its nat-
ure that affects the structural transformation process
(Christiaensen, De Weerdt, & Kanbur, 2016; Ferré, Ferreira, &
⇑ Corresponding author. Lanjouw, 2012). Urbanization in Africa is characterized by a con-
E-mail addresses: [email protected] (J. Vandercasteelen), centration of individuals in smaller urban centers. Only 10 percent
[email protected] (S.T. Beyene), [email protected] of the African urban population resides in larger cities with
(B. Minten), [email protected] (J. Swinnen).
https://doi.org/10.1016/j.worlddev.2018.03.006
0305-750X/Ó 2018 The Authors. 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/).
394 J. Vandercasteelen et al. / World Development 106 (2018) 393–406
between 5 million and 10 million inhabitants—the majority of the Ethiopia, our analysis and empirical results show that the size of
urban population lives in medium or small-sized cities of between the city matters for agricultural prices and intensification. There-
1 million and 5 million inhabitants (35 percent) or in small urban fore, these results suggest the importance of differentiating city
areas (55 percent) (UN Population Division, 2014). Moreover, the size when estimating the impact of urbanization on agricultural
populations in medium-sized cities has doubled in the last decade transformation, both in empirical regressions and in computable
and is expected to grow by more than 30 percent in the next dec- general equilibrium (CGE) models.
ade (UN Population Division, 2014). As a consequence, the urban Second, this paper also relates to the literature that examines
population in Africa is widely dispersed across cities of different the effect of transportation costs on different aspects of economic
sizes. growth in rural areas (Jacoby & Minten, 2009; Gollin & Rogerson,
Urbanization affects poverty levels in the rural hinterland of 2014; Jedwab & Moradi, 2016; Storeygard, 2016). We contribute
urban cities through multiple spillovers and economic linkages to to this literature in two important ways. First, while most research
these rural areas. Migration to urban centers is associated with uses distances to cities to identify urban proximity, we calculate
cash flows towards the rural hinterland (e.g. remittances), and for each farmer the transportation costs to potential market out-
the growth of secondary towns also indirectly affects rural poverty lets. Second, we control for the potential endogeneity of trans-
levels through upward pressure on rural wages, and rural non- portation costs by using an IV-identification strategy that
farm income opportunities (Cali & Menon, 2013; Dorosh & introduces exogenous variation in transportation costs using geo-
Thurlow, 2013). Moreover, consumption linkages between urban referenced data to estimate natural path transportation costs
markets and rural producers seem important, primarily because (Damania et al., 2016). We illustrate important effects of trans-
higher food consumption and changing diet preferences in urban portation costs to primate and secondary cities on staple crop
centers increase urban demand for rural agricultural products prices and production practices through different econometric
(Djurfeldt, 2015; Dorosh & Thurlow, 2014; Reardon & Timmer, methods, with a battery of controls of household and farm charac-
2014; Tschirley, Haggblade, & Reardon, 2013). However, it remains teristics, as well as controls for self-selection and endogeneity of
unclear which type of urbanization (metropolization vs. more dis- transportation costs.
persed) is most beneficial for farmers in the rural hinterland and
through which channels (different sized) cities affect rural produc-
ers (Gibson, Datt, Murgai, & Ravaillion, 2017). 2. Urbanization patterns and teff production in Ethiopia
While there is no general consensus on how urbanization pat-
terns affect rural livelihoods, a hierarchical pattern of settlements In Ethiopia, the population living in cities is expected to grow
is believed to be more conducive to modernize and commercialize from 15.2 million in 2012 to 42.3 million by 2034, corresponding
subsistence agriculture (Brutzkus, 1975).1 As smaller cities are to an annual growth of 5.4 percent (World Bank, 2015). Addis
more closely located to the rural hinterland, the production and mar- Ababa is by far the largest city in Ethiopia and about a quarter of
keting linkages for agricultural products could be stronger because the urban population in Ethiopia lives in the capital (Schmidt &
of lower transportation costs and stronger local ties (Richards Kedir, 2009; CSA (Central Statistical Agency), 2013b). At the same
et al., 2016). Hence, the growth of secondary towns could directly time, the expansion of smaller and medium sized cities is on the
benefit rural farmers – and other actors in agricultural value chains rise. This affects urban – rural relationships. For example, the per-
– because of improved market access and opportunities (Reardon, centage of the rural population less than 3 h away from a city with
2016). Moreover, secondary towns provide rural farming households a population of at least 50,000 has increased from 15 percent in
access to specialized services and facilities, input markets, and non- 1997/1998 to 47 percent in 2010/11, partly driven by this city
farm employment opportunities (Richards et al., 2016). Furthermore, growth, but also by infrastructure improvements (Kedir, Schmidt,
as cultural ties and social networks might be stronger in smaller & Tilahun, 2015).
sized towns, these might be more effective in spreading and dissem- In our analysis we focus on how these cities are related to teff
inating new ideas, agricultural innovations, and knowledge to farm- prices and production in Ethiopia. In 2011, teff constituted 23 per-
ers in the rural hinterland (Brutzkus, 1975; Rondinelli, 1983; cent of the total grain crop area and 17 percent of total grain pro-
Berdegué, Carriazo, Jara, Modrego, & Soloaga, 2015). duction in Ethiopia (CSA (Central Statistical Agency), 2012). 29
This paper contributes to the literature on secondary towns and percent of teff production is sold, which is a relatively high share
on spatial economics in the rural hinterland. First, we investigate compared to other cereals, such as wheat and maize (at 20 percent
the role of city size in the rural development process and 11 percent, respectively). Hence, teff has a higher commercial
(Christiaensen, Demery, & Kuhl, 2011; Christiaensen, Weerdt, & surplus, and is often considered a cash crop for farmers engaged in
Todo, 2013; Christiaensen & Todo, 2014; Dorosh & Thurlow, its production (Minten, Tamru, Engida, & Kuma, 2015, 2016). The
2013, 2014; Berdegué, Carriazo, Jara, Modrego, & Soloaga, 2015; production of teff in the major production areas is increasingly
Gibson, Datt, Murgai, & Ravaillion, 2017; Ingelaere, Christiaensen, coordinated in a local value chain, and especially so in the produc-
De Weerdt, & Kanbur, 2018). Unlike the previous literature – which tion areas around the capital (Minten et al., 2016). Teff farmers
has mostly focused on poverty outcomes for migrants (i.e. struc- receive a large share of the final teff price, which ranges around
tural transformation) – we explore how urbanization patterns 80 percent depending on the quality traits of the teff sold and
affect output prices and agricultural practices of farmers in the the type of output market (Minten et al., 2016). Using the same
rural areas (i.e. agricultural transformation). We first develop a dataset as used in this paper, Vandercasteelen, Tamru, Minten,
conceptual framework illustrating the effect of different sized and Swinnen (2018b) document that net teff profitability (mone-
towns on agricultural prices and intensification outcomes. Using tary value of teff production minus input expenditure – including
a unique large-scale survey of staple crop (teff) producers in labor) for farmers in the commercial production zones around
Addis Ababa is on average 410 USD per hectare in 2011. Similarly,
Vandercasteelen, Dereje, Minten, and Taffesse (2018a) find that the
1
In contrast, concentration in primate cities (e.g., national capitals) increases the total net returns in ETB per hectare of land is 385 USD per hectare
demand for agricultural products, provides economies of scale for commerce at large in Oromia. More disaggregated data on teff prices, profits and pro-
central markets and concentrates the development of new agricultural technologies
and innovations (Brutzkus, 1975). Therefore, spread effects are hypothesized to be
ducer margin for different zones is presented later on in Table 1.
stronger for larger cities and for farmers located in the close vicinity of such cities Teff is more readily eaten in urban than rural areas. In urban
(Benziger, 1996). areas, teff has a high share (23 percent) of per capita consumption
J. Vandercasteelen et al. / World Development 106 (2018) 393–406 395
in total food consumption (Berhane, Paulos, & Tafere, 2011). In improved seeds, as agricultural intensification usually involves the
Addis Ababa, teff accounts for almost half of total cereal expendi- increased uptake of these inputs (see Vandercasteelen, Tamru,
ture, and teff is consumed more by richer and urban households. Minten, and Swinnen (2018b) for an analysis of land and labor use).
The income elasticity of demand for teff in urban areas is around The nominal prices of fertilizer and improved seeds are unlikely to
1.1 (Berhane et al., 2011). Because cities and urban incomes are change over distance in Ethiopia, as they are sold by state-controlled
growing quickly, the demand for teff is expanding and expected cooperatives at fixed prices in each village (Rashid, Tefera, Minot, &
to further increase (Minten et al., 2015, 2016). Ayele, 2013). However, while the nominal prices may be the same
As shown above, teff is especially relevant for Ethiopian policy for all cooperatives, Minten, Koru, and Stifel (2013) find that the trans-
makers because of its food security and income consequences for action costs of obtaining fertilizer and improved seeds are much
rural producers and as an important food for urban consumers. Teff higher in remote areas.5 If we interpret rL as the per unit opportunity
production and consumption is mainly restricted to Ethiopia, and cost of acquiring fertilizer and seeds, then @r
@d
L
> 0.
the value of domestic consumption largely exceeds the value of The farmer maximizes the following profit function
the international trade of teff (mainly in the form of injera).2 Teff
marketing in Ethiopia provides therefore a closed economy setting maxP ¼ maxðp Y rL LÞ ð3Þ
L
and, as a consequence, teff prices – as well as other intensification
outcomes – are not directly determined by international prices. Teff which yields the input demand function:
is therefore an interesting case to study domestic urbanization
PL : pðdÞ ½AðdÞ f L ðL Þ rL ðdÞ ¼ 0 ð4Þ
induced agricultural transformation in rural hinterlands.
This first order condition (FOC) defines the optimal demand for
3. Urbanization and agricultural intensification: A conceptual input L as a function of the distance to the city (d):
framework
@L @rL @p @A
¼ þ A f L hL þ p f L hL ð5Þ
@d @d @d @d
3.1. One city model
where hL ¼ 1=½1 p A f LL . Eq. (5) shows that there are three
Consider an agricultural commodity that is produced by a rep- effects of distance from the city on the input used on a farm. The
resentative farmer in a rural area and sold in a city (City 1) where first term captures the own price effect of the input – in the case
all the consumers live. Define d as the distance from the farm to the of modern seeds and chemical fertilizer, this is either zero or nega-
city. The farmer uses input L to produce agricultural output Y. The tive. The second term and third term, respectively, capture the out-
production function is put price effect and the impact of declining productivity with
Y ¼ AðdÞ f ðLÞ ð1Þ distance. With hL positive, the output price effect (second term)
and the productivity effect (third term) are definitely negative: a fall
where f ðLÞ has standard concave properties of f L > 0 and f LL < 0. A(d) in the output price and in productivity both reduces the demand for
is a factor neutral productivity shifter – which is assumed to cap- inputs. Hence, our simple model predicts that the effect of distance
ture the direct productivity effect of being close to the city – with on the uptake of improved seeds and chemical fertilizer (or other
@AðdÞ
@d
< 0.3 The effective price the farmer receives for his output is a modern inputs) is likely to be negative.6
function of the price in the city and the distance to the city Fig. 1 presents a very simple graphical illustration of our ‘‘one-
(Minten & Kyle, 1999). The price of the agricultural commodity in city-model”. City 1 (‘‘the capital”) is located at point 0 on the hor-
the urban market of City 1 is pu . Define lðdÞ as the per unit cost of izontal axis. All farmers are located to the right of City 1 on a single
transportation to the city with @ l@dðdÞ > 0. The effective output price distance line, with distance represented by transportation costs m.
for the farmer, p, is then The vertical axis measures the size of I (which is an indicator) that
captures key variables of interest (prices and the use of modern
p ¼ pu lðdÞ ð2Þ inputs). Our model predicts that all these variables (prices and
Distance to the city may also affect the costs of inputs for farmers. modern input use) will decline with distance to the city, repre-
As a general specification, we can write the cost per unit of input t sented by a declining I(m) function. Obviously the shape of the
rL ¼ rL ðruL ; dÞ, with ruL the price of input L in the urban market. The function (linear or not), its slope, and the height of the intercept
will depend on the specific variable (prices or various inputs).
impact of d on rL ð@r @d
L
) is likely to depend on the nature of the input.4
However, as we will show further, for illustrative purposes this
We focus here on the use of modern inputs, i.e., chemical fertilizer and
simplified general function is a useful representation when we
add complexities, such as by adding more cities to the model,
2
In 2015, the total teff production (in both rainy seasons) was estimated at and when we feed empirical data into the model at a later stage.
4,576,900 metric ton (CSA, 2015/16). Valuing this production – assuming the average
retail price of Addis Ababa (which was about 15,510 ETB per metric ton) and the
annual average exchange rate (USD = ETB) of 21.1059 – gives a total value of teff 3.2. Secondary town model
production of about 3.4 Billion USD. Data from the Ethiopian Custom’s Authority
indicates that the value of injera export has increased from 1.7 million USD in 2008 to
We now extend this framework by analyzing how the presence
about 10.2 million USD in 2015 (Minten et al., 2015). However, the share of injera
export (through formal route) amounts to only 0.3% of the total value of teff of more (secondary) towns changes these relationships. With more
production in 2015. cities, we can define an I-function for each city i which represent
3
The productivity effect comes from better access to infrastructure and services, how prices and modern input use will change with distance from
extension and information, improved technologies and better networks for farmers city i, i.e., Ii(mi) with mi transport costs to City i.
closer to cities (Stifel & Minten, 2008, 2017; Josephson, Ricker-Gilbert, & Florax, 2014;
Vandercasteelen et al., 2018b).
4 5
If the farmer buys his inputs in the city and has to transport this to the farm, his One factor is the uncertainty of suppliers; hence farmers have to go multiple
input costs will increase with distance to the city. However, if the city competes with times to the cooperative to acquire their inputs. Minten et al. (2013) show a
the farm as a potential alternative use for the input, there will be opposing effects of significant positive effect of distance to a market town on the time it takes to acquire
distance on the input price because of increased competition and at the same time inputs.
6
lower transport costs. This is particularly relevant for labor input. If the alternative For the case of other inputs (labor or land) in the one input case, or for the case of
employment for labor is employment in the city, the price of labor will be lower the multiple inputs (there may be additional substitution or complementarity effects),
further away from the city. see Vandercasteelen et al. (2018b).
Fig. 1. Impact of proximity to City 1 on I (prices and modern input use).
Fig. 2. Impact of capital City 1 and secondary City 2 on I (prices and modern input
A key question relates to differences in the shape of the I- use).
functions for different cities. In terms of the linear I-function of
Fig. 1, this means whether the slopes of the I-functions are the of modern inputs by farmers. If there were no secondary town
same and whether they have the same maximum. One can imagine effects, we would find a continuously declining relationship
different reasons why neither the maximum value nor the slope between prices or modern input use and transport costs from the
should be identical for different cities. For example, in the case of primate city. If the presence of the secondary town does influence
agricultural output prices, it may be that the price in (smaller) sec- prices and agricultural intensification measures, we expect to see
ondary towns is lower than the price in the capital, since demand the decline of the primate city’s impact disrupted by increasing
may be lower in smaller cities, transaction costs may be lower, or prices and input use due to impact of secondary cities, as illus-
markets may not be as well integrated. In this case, the maximum trated in Figs. 2 and 3. In the following sections, we will first sim-
value Ii(0) would differ. It is also possible that per unit costs of ulate the predicted impacts of our conceptual model on output
transport are higher closer to smaller secondary towns because prices using actual transport cost data, and compare these simu-
of poorer road infrastructure or because of less competition in lated results with actual observations. Later, we will estimate the
the transport sector (possibly linked to thinner markets). In this impact of the secondary cities using econometric techniques.
case the slope of the Ii(mi) would be steeper. These are essentially
empirical issues. For our conceptual framework, we will illustrate
the impacts under different assumptions. 4. Data and descriptive statistics
Consider first the case where there is one additional (secondary)
city (City 2) with the same slope of the I-function as City 1 but a We use data on teff producers from Ethiopia collected in
lower maximum I value (I1(0) > I2(0)). This case is illustrated by November and December 2012. Appendix I explains the sampling
Fig. 2. Farmers are located to the right of the capital (City 1) and procedure that was used. We focus on the zones in this dataset that
at both sides of the secondary town (City 2). The black line repre- contain secondary towns of substantial size, at least in terms of
sents the I1 function, which is the impact of City 1 on prices or population numbers, that function as an important regional hub
modern input use, and the gray line represents the I2 function, or consumption center of teff in Ethiopia. Nazareth (or Adama) is
which is the impact of City 2. Farmers which are closer (in terms located in East Showa (southeast of Addis Ababa) and is an impor-
of transport costs) to City 1 than m⁄1 will be dominated by the influ- tant hub from which to export teff to the regional capitals in the
ence of City 1 (as I1(m1) > I2(m2)). Farmers which are further away east (Dire Dawa or Harar) and the south (Hawassa) of Ethiopia.
from City 1 (to the right of m⁄1 on the vertical axis) will be influ- Bahir Dar is located in West Gojjam (northwest of Addis Ababa)
enced mostly by City 2, as the proximity to secondary City 2 will and is an important hub for transporting teff to the cities of
determine their prices and modern input use, rather than distance Mekelle and Gonder in the north. We also assume that (some)
to the capital city (as I1(m1) < I2(m2) for these farmers). It is easy to farmers in East Gojjam might be affected by Bahir Dar, as there is
see that the lower I2(0), the smaller the impact of the secondary no secondary town of substantial population numbers nearby. As
city (and vice versa). In fact, if I2(0) would be less than I1(m21), there a consequence, the empirical analysis will focus on the 720 farmers
would be no effect of the secondary city. Similarly, the impact of who are located in the three zones of West Gojjam, East Gojjam
the secondary city would be smaller if the slope of the I2 function and East Showa.7
is steeper, as higher transport costs would reduce its impact region The map in Fig. 4 gives an overview of the survey areas and the
(and vice versa). location of the farmers that were interviewed. It is shown that we
Fig. 3 presents a more complex example where the capital City have data on farmers who are more or less located along a straight
1 is in the center and there are secondary cities on both sides. The line from Bahir Dar to Nazareth, passing Addis Ababa and con-
‘‘impact region” of City 1 is now the transport costs region starting nected by one of the most important national roads in Ethiopia.
⁄
from m# 1 to the left of City 1 to m1 to the right. All farmers who are
within this distance are influenced by City 1 in the prices they set 7
Nazareth is projected to become the second largest city in Ethiopia in 2016
and their input use. Farmers who are further away are influenced (342,940 inhabitants), while Bahir Dar will be the sixth largest city with 297,794
by secondary cities, either City 2 on the right hand side (farmers inhabitants. Based on the international standards, these cities would be classified as
for which m1 < m⁄1) or by City 3 on the left hand side (farmers for ‘‘Urban areas smaller than 500 000” (UN Population Division, 2014). There is no
sizable secondary town in the vicinity of the surveyed farmers in South West Showa
which m1 < m# 1 ). (the fourth zone in our dataset). The largest city in West Showa (the fifth zone) is
This simple conceptual model suggests hypotheses on how the Ambo, which is however small (less than 100,000 inhabitants) and the surveyed
presence of secondary towns affects agricultural prices and the use farmers in that zone are not located along the route from Ambo to Addis Ababa.
Fig. 3. Impact of capital City 1 and secondary Cities 2 and 3 on I (prices and modern input use).
To make the link with the conceptual model, we consider Addis cities and we select for each trader market the city that maximizes
Ababa to be at the origin, while farmers in East and West Gojjam trader margins. This city will be used as the most likely destination
are located on the left of Addis Ababa and farmers in East Showa of teff, and is either Addis Ababa or one of the secondary towns.10
are located on the right of the capital. Hence, the data suits the We will refer further on to farmers whose teff is shipped teff to Addis
general specification of the two-dimensional model of farmers Ababa as ‘Addis Ababa farmers’ and those whose teff is shipped to a
located on the straight line that connects three cities, with the pri- secondary town as ‘secondary town farmers’.
mate city, Addis Ababa, in the middle.8 The first panel of Table 1 provides an overview of the different
Our main variables of interest are the final urban markets (i.e. prices (urban and farmer level), transport cost (from farm to trader
cities) where farmers’ teff is sold and how far farmers are located town and from the trader town to the city), trader margins, teff
from that city. We consider the above three cities (Addis Ababa, profits and the number of farmers in our survey who supply to
Nazareth and Bahir Dar) as the major regional teff markets. To each of the three cities. The data is presented for Addis Ababa
identify – for each farmer – the most likely city where teff is and secondary town farmers separately, where the latter is further
shipped to, we use the assumption that traders will ship teff from divided into Bahir Dar and Nazareth. Teff trader margins range
the local trader market to the city that maximizes the net trader from 78 to 182 ETB per quintal, and farmers receive a share of
price of teff. Unfortunately, we have no data on prices traders about 80 percent in the final retail price. Both numbers are in line
receive when selling teff in urban markets in Ethiopia. Therefore, with margins and shares reported in Minten et al. (2016). Net prof-
we calculate the trader margin for different potential urban mar- itability of teff producers is highest for those farmers whose teff is
kets as the difference between the retail price in each of the urban shipped to Addis Ababa, presumably because they have higher
market and the price traders paid to farmers. Then, the trader mar- yields (results not reported). The second panel of Table 1 also
gin is calculated net of the total cost of shipping teff to the city shows the descriptive statistics on the household head and farm
incurred by traders.9 The net trader margin is calculated for all three characteristics of Addis Ababa and secondary town teff farmers.
On average, these farmers are similar in characteristics listed in
Table 1, although secondary town farmers seem to be slightly older
8
The geographical location of the secondary towns in space determines how far the and have larger families (but only significant at the 10 percent
influence of the secondary town on the farmers reaches, relative to farmers’ distance level).
from Addis Ababa. For example, most of the farmers in West and East Gojjam – with a
We are further interested in capturing the effect of urban dis-
few exceptions – are located between Bahir Dar and Addis Ababa. For these farmers,
the effect of the secondary town is likely to be more visible as both cities are located tance on agricultural outcomes. We use the transportation costs
quite far from each other. On the contrary, Addis Ababa and Nazareth are located that farmers incur when their teff is shipped to a city. Such cost
close by, and there are farmers located on both sides of Nazareth. Hence, the influence indicator is assumed a better measurement of market proximity
of each city is determined by the prices of teff in each city, the distance of each farmer
than physical distance or travel times (Chamberlin & Jayne,
with respect to the capital and secondary town, and the transportation cost functions
for each city. Defining this rural hinterland corresponds with determining – for each
farmer – the most profitable city for which to produce.
9 10
Urban teff (retail) prices in the markets of Addis Ababa, Nazareth, and Bahir Dar If the net price margin the traders would receive from shipping teff to Addis is
come from the Central Statistical Agency during the survey period (CSA (Central higher than the net price margin in Bahir Dar, then the trader is assumed to be selling
Statistical Agency), 2013a). The calculation of the shipping cost that traders incur is teff in the urban market ofAddis Ababa and vice versa. For example, for some local
explained in the next paragraph. To this cost, an additional cost is added to capture trader markets in the region of West Gojjam, the end destination of teff will be Bahir
transport brokers’ fee as the majority of traders use brokers to find a teff load (Minten Dar, because this city maximizes trader price margins; while for other trader markets
et al., 2016). Note that this trader margin does not take into account transportation in East Showa the urban market Nazareth provides the highest price margin. The
costs that farmers themselves incur as farmers have no influence on where traders different colors for farmers in Fig. 4 correspond to the city where they are the most
sell teff. likely to ship their teff to.
Fig. 4. Overview of the survey area with major roads and the three studied cities (squares). Note: The surveyed farmers (dots) are colored according to the city they supply to.
Table 1
where the teff is most likely to end up. This cost of shipping teff is
Classification and end destination of farmers. the combination of two separate costs. First, we use farmers’ self-
reported cost incurred in travelling with their harvested teff from
Variabele End destination of farmers
their farm to the local market. As donkeys are the most common
Addis Bahir Nazareth mode to transport teff to the trader town, this can be considered
Ababa Dar
as a ‘pack animal cost’. Second, in these local markets, traders pur-
Zone Addis East East chase teff and ship it to one of these three cities. As the teff is
Ababa Gojjam Showa
shipped from a local market to a city by motorized trucks, we
Teff price in city (ETB/quintal) 1,364 1,335 1,344
Teff trader margin (ETB/quintal) 182 171 78 use a spatial network analysis function in GIS to calculate the
Transport cost of shipping teff from trader 43 29 20 ‘truck cost’ for the second trip. The exact details on the calculations
market to city (ETB/quintal) of transport costs are explained in Appendix II. These results were
Transport cost of transporting teff from home 12 15 18
calibrated with data on transportation costs from community sur-
to trader market (ETB/quintal)
Price received by farmer (ETB/quintal) 1,073 1,057 1,158
veys and from a truck driver survey implemented at the same time
Share of farmers’ price in urban price (%) 79 79 86 as the teff producer survey. Table 2 provides an overview of the
Net profits (ETB/ha) 11,621 6,970 8,122 transportation cost measures and other variables of interest. On
Farmers (number) 177 359 184 average, it costs 47 ETB per quintal to ship teff from the farm to
Farmers (%) 25 50 25
the city, of which the ‘pack animal cost’ contributes about 13 ETB
Variabele Classification of farmers
per quintal and the truck trip from local market to the city costs
Addis Secondary Town
on average 34 ETB per quintal. This cost is in line with estimates
Ababa
of transport cost of other surveys on the teff value chain (Minten
Age of head (years) 44 46*
et al., 2016).
Gender of head (male = 1) 95 95
Educated head (years = 1) 51 52 On top of prices and costs, different intensification outcomes in
Ethnicity (Oromia = 1) 26 28 teff production are also reported in Table 2 for the selected 720
Size of the household (number) 6 6* farmers. These outcomes are input (fertilizer and improved seeds)
Household owns mobile phone (yes = 1) 34 25 application and teff land productivity (kg per ha).11 Such modern
Farm assets (ln of ETB) 6 7
inputs as these have been promoted by the government to stimulate
Travel time to nearest dry weather road 29 35
(minutes) agricultural production in the country (Bachewe et al., 2018). During
Travel time to nearest all weather road 58 67 the household survey, detailed data on production practices were
(minutes) gathered, as well as on inputs applied and on teff output for each teff
Household took loan last year (yes = 1) 34 31
plot level. This was then averaged over all plots cultivated by the
Household is member of agricultural 55 65
cooperative (yes = 1) household. Further, data on teff prices (ETB per quintal) were col-
lected in the household survey (teff transaction section) which
Note: The asterisk in the second panel indicate whether the comparison of the
contained information on quantities sold, prices received, main place
means of the two groups of farmers yielded a significant difference at the 10%.
Number of observation is 720.
11
All the intensification outcomes are trimmed (replacing the top and bottom 1
2013). In practice, this transportation cost is calculated as the cost percent of non-missing values by village average) to control for potential outliers.
of transporting one quintal (100 kg) of teff from the farm to the city Results are similar if the data is not trimmed.
Table 2
Descriptive statistics.
Variable Unit Data source Mean Median SD

Prices
Price of teff ETB/quintal Teff transaction level 1,035 1,027 116
Wage rates ETB/day Community level 37 38 12
Land rental rate ETB/ha Community level 4,697 4,709 130
Price of DAP ETB/quintal Community level 1,384 1,411 115
Price of Urea ETB/quintal Community level 1,128 1,167 120
Agricultural inputs
Use of DAP kg/ha Farmer plot level 101 96 78
Use of Urea kg/ha Farmer plot level 71 51 76
Use of improved seeds kg/ha Farmer plot level 13 0 22
Intensification outcomes
Teff land productivity kg/ha Farmer plot level 1,231 1,179 671
Teff profits ETB/ha Farm plot level 8,408 7,468 5,254
Transportation cost measurements
Pack animal cost ETB/quintal Teff transaction level 14 14 8
Truck cost ETB/quintal Trader and truck survey, road network 33 33 19
Total transportation cost ETB/quintal Pack animal and truck cost 47 47 16
Natural path Hours Own calculation 17 15 10
Note: Number of observations in each model is 720. ‘SD’ is the standard deviation.
of sales, and the buyer in each teff sales transaction during the last
production season. Monthly village level wage data were further
aggregated to yearly averages (in ETB per day).
5. Simulation results and observed teff prices
The theoretical model shows that if secondary towns influence

rural teff producers, there should be nonlinearities in the relation-
ship between intensification outcomes (e.g. agricultural prices and
modern input use) and the distance to Addis Ababa. As a first step,
we use retail price data as a benchmark to simulate the model pre-
dictions for effective teff prices farmer receive (i.e. Eq. (2)). Fig. 5
plots simulated (Graph 1) and actual prices – reported by farmers
– (Graph 2) of white teff against the truck cost to ship teff to Addis
Ababa.12 To ease interpretation, we posit farmers in East and West
Gojjam on the left of the origin by assigning them negative trans-
portation costs. This allows to compare the model predictions illus-
trated in Fig. 3 with the empirical relationship of prices and
proximity to the capital Addis Ababa.13
The upper graph of Fig. 5 Figure illustrates how the distance to
Addis Ababa affects the teff price net of transport cost farmers in
the rural hinterland of Addis Ababa receive (as simulated using
the model in our conceptual framework). In the lower graph of
Fig. 5, the empirical relationship between reported teff prices and
distance to Addis Ababa is presented. In general, we observe that
the predicted prices and prices reported by farmers follow a similar
trend over transportation costs to Addis. Farmers’ teff price in East
and West Gojjam (on the negative part of the x-axis) decreases
with distance, up to a certain transport cost where the effect of
Fig. 5. Predicted (above) and observed (below) teff price over transport cost to
the presence of Bahir Dar appears. At this point (i.e. m1 in Fig. 3), Addis Ababa. Note: The solid lines represent local polynomial smoothing estimates
the net price that farmers receive in Bahir Dar becomes higher than of transportation cost (x-axis) on output prices (y-axis). The shaded area
the net price in Addis and teff will be shipped to Bahir Dar. The teff corresponds with the 95% confidence intervals.
price for farmers with higher transport cost will be positively
affected by the distance to Addis Ababa. This negative relationship
with distance to Addis therefore holds up to a kink, where the sec- the location of Nazareth. For farmers located further than Nazar-
ondary town is located, and afterwards prices are again negatively eth, prices are again strongly and negatively related to distance
related to distance to Addis. The net price received by farmers on from Addis. Hence, the fact that there are nonlinearities in both
the right of Addis Ababa shows also a kinked pattern. With low (left and right) price relationships with respect to distance to Addis
transport costs, net teff prices decrease with distance up to a cer- Ababa suggests that secondary towns indeed affect teff prices.
tain transport cost where prices become more or less flat, i.e. at The effect of secondary towns on teff intensification outcomes
is not only present in teff output prices. In Fig. 6, we present
12
White teff is used as this type of teff is most commonly traded in Ethiopia.
non-parametric regression results on land intensification measures
13
Given that the truck cost is calculated using the road segments, this measurement in function of transport cost to Addis Ababa. The figures show how
takes into account differences in road quality and thus transaction costs. fertilizer use (Graph 1, for Diammonium phosphate (DAP)) and
soils, and the share of flat – versus sloped – land). Moreover, prices
of outputs and inputs are additionally included for the models where
intensification measures are the dependent variables.16 We further
also include zonal fixed effects (Zi ) as an additional control for farm-
ers who are in the same geographical zone. Standard errors are clus-
tered at the village level.
There are two main estimation issues in this empirical setting.
First, there might be selection bias because of systematic differ-
ences between Addis Ababa farmers and those supplying sec-
ondary towns. This might be especially relevant for teff
production, as teff is differentiated by color and origin, and con-
sumers in urban centers generally prefer and are more willing to
pay for higher quality teff. As a response, Addis Ababa farmers
might adjust teff production and focus on the production of higher
quality teff. We address this potential selection bias by first con-
trolling for observational differences, and then by comparing Addis
Ababa and secondary town farmers that are similar in these
observable characteristics.
We therefore apply a double robust regression approach to con-
trol for observable selection issues. First, we estimate the probabil-
ity to be a secondary town farmer based on household
characteristics (age, gender, education, ethnicity, household size,
share of non-farm income), asset ownership (radio, TV, mobile
phone, wealth index), and agricultural characteristics (application
for loan, membership in and distance to cooperatives, value of agri-
cultural assets, total cultivated teff land owned by the house-
hold).17 We further add a proxy for teff farming ability that is
estimated from a panel fixed effects model of a teff production at
the plot level.18 This proxy will control for observational differences
Fig. 6. Usage of DAP (kg/ha) and Improved Seeds (kg/ha) over transport cost to in farm capabilities between Addis Ababa and secondary town farm-
Addis Ababa. Note: The solid lines represent local polynomial smoothing estimates ers. Finally, we add the share of teff production that is allocated to
of transportation cost (x-axis) on different teff outcomes (y-axis). The shaded area
higher quality varieties to capture the potential specialization by
corresponds with the 95% confidence intervals.
Addis Ababa farmers on higher quality teff varieties (white or
magna).19 From this probit model, we then estimate the propensity
improved seed (Graph 2) use change with transportation costs, score and the results of this regression are reported in Table 3.
similar to Fig. 5. We see that the use of both inputs does not lin- Second, we use the inverse of the propensity score to weight our
early decrease over space, and is therefore seemingly affected by regressions in the subsequent analysis to control for potential
the presence of the secondary towns Bahir Dar and Nazareth.14 selection biases in Si .20
These graphs thus tend to suggest that secondary towns do affect Second, it can be argued that roads (transport infrastructure)
intensification in the rural hinterland in Ethiopia. are not placed randomly in space and are denser in areas with
higher economic potential. Hence, our measure of transportation
6. Regression analysis cost could be endogenous. We therefore apply an instrumental
variable (IV) approach to address this concern. The implemented
6.1. Empirical specification approach follows the methodology pioneered by Damania et al.
(2016). We instrument transportation costs with a ‘natural path’
To measure the relationship between types of cities and prox- measure, which is the time (in hours) it takes to walk from a pro-
imity to a city on the one hand, and agricultural prices and inten- duction area to the market place, in the absence of any transporta-
sification on the other hand, we estimate the following regression tion infrastructure. This measure takes into account the effect of
model: geography, i.e., travel speed adjusted by slope, and does not follow
the potentially endogenous road networks in areas with higher
Yi ¼ ay þ by Ti þ cy Si þ Xi þ Zi þ ei;y ð6Þ economic potential. Instead, it provides the most efficient, i.e.,
where Yi is the outcome variable of interest, T i is the total transport
cost from the farm to the city where the teff is shipped to15, Si indi-
16
cates whether a farmer is a secondary town farmer compared to an To do so, the reported land rental rates, input and teff output prices are regressed
Addis Ababa farmer, ei;y is the idiosyncratic error term of outcome on the parcel-level and transaction-level determinants of prices respectively using a
fixed effect model. From these panel estimations, the predicted values of the land
Yi , Xi a matrix of household controls and contains (i) age, gender, rental, output and input price are calculated for each farmer and are used as
ethnicity, and education of the household head, (ii) household’s independent variables in the implemented regressions.
membership in an agricultural cooperative, assets value, land hold- 17
The wealth index is constructed using a Principal Component Analysis on
ings, ownership of mobile phone/radio and member size; and (iii) different housing assets.
18
agro-ecological conditions (altitude, the share of brown or black For more information on the construction of the proxy for teff farming ability, we
refer to Vandercasteelen et al. (2018b).
19
The wealth index is constructed using a Principal Component Analysis on
14
Teff yields also show nonlinearities at both sides of the capital, but the effect of different housing assets. The farming ability in teff is proxied by the household fixed
the secondary towns (i.e. the kinks) are less clear as for teff prices and improved seed effects from estimating a teff production function.
20
usage (results not reported). Secondary town farmers are weighted by the inverse of the propensity score,
15
This measure of distance is not to be confused with the distance to the capital while Addis Ababa farmers are weighted by the inverse of 1 minus the propensity
Addis Abeba used in the previous section. score.
Table 3 Table 4
Estimation of propensity score. First stage regression.
Explanatory variables Secondary town farmer First stage regression Transportation Cost
(Yes = 1) (ETB/quintal)
Coefficient Standard Error Coefficient Standard
Error
Age of head (years) 0.002 (0.005)
Gender of head (male = 1) 0.061 (0.236) Natural Path (hours) 1.58*** (0.096)
Educated head (years = 1) 0.226** (0.104) Secondary Town 3.05* (1.56)
Head is from Oromia (yes = 1) 0.017 (0.392) Teff price (ETB/qnt) 11.3*** (2.48)
Size of the household (number) 0.006 (0.042) Wage (ETB/day) 1.08 (1.67)
Household owns radio (yes = 1) 0.022 (0.125) Land rental rate 61.5*** (18.1)
Household owns tv (yes = 1) 0.964*** (0.312) Price of DAP 2.38 (8.12)
Household owns mobile phone (yes = 1) 0.358** (0.152) Price of Urea 20.5*** (6.25)
Farm assets (ln of ETB) 0.198** (0.086) Age of head (years) 0.13*** (0.036)
Household applied for loan (yes = 1) 0.095 (0.100) Gender of head (male = 1) 2.89 (2.07)
Household is member of cooperative (yes = 1) 0.245 (0.180) Educated head (years = 1) 0.22 (0.92)
Share of non-farm income (%) 0.009 (0.007) Head is from Oromia (yes = 1) 0.46 (1.90)
Distance to agricultural cooperative (minutes) 0.004 (0.002) Household is member of agricultural cooperative 0.19 (0.94)
Wealth index (PCA of housing assets) 0.351 (0.215) (yes = 1)
Farming Ability (.) 0.568*** (0.181) Farm assets (ln of ETB) 0.70*** (0.23)
Land owned (ha) 0.175*** (0.060) Size of the household (number) 0.091 (0.21)
Share of teff land allocated to high quality (%) 0.006** (0.003) Land owned by the household (ha) 1.22***
(0.38)
Constant 0.355 (0.820) Household owns mobile phone (yes = 1) 2.31* (1.25)
Household owns radio (yes = 1) 1.79 (1.21)
Note: Number of observations is 720. Standard errors are clustered at village level Altitude (m) 0.014*** (0.0026)
and reported in parentheses next to the coefficient: ***p < 0.01, **p < 0.05, *p < 0.1. Share of black and brown soils (%) 1.97* (1.03)
For definition of the ‘Farming Ability’ variable, see Vandercasteelen et al. (2018b). Share of flat soils (%) 3.17** (1.26)
***
Constant 510 (152)
Zone Fixed Effects yes
the least costly in terms of hiking time, path that farmers would
Observations 720
take if they had to transport their teff on foot. R-squared 0.63
Eq. (6) is estimated both using Ordinary Least Squares (OLS) and Adjusted R-squared 0.62
Two-Stage Least Squares (2SLS), where we instrument transport Shea’s Partial R-squared 0.28
cost with the ‘natural path’ measure to control for potential Robust F(1,697) 273***
endogenous effects. The results of the first stage regression for Note: Number of observations is 720. Standard errors are clustered at village level
the 2SLS method are reported in Table 4. It shows that an addi- and reported in parentheses next to the coefficient: ***p < 0.01, **p < 0.05, *p < 0.1.
tional hour of walking from the farm to the city where the teff is
sold increases – all else being equal – transportation costs by 1.5
city that is the end destination of teff, has a significant and negative
ETB per quintal. The bottom of Table 4 reports the results of
effect on the output price in all regressions.
first-stage tests for weak instruments. The different R-squared
The plot level usage of modern inputs per hectare is shown to
measures (normal, partial, and adjusted) indicate a strong correla-
be significantly lower for secondary town farmers, compared to
tion between the natural path measure and transportation cost.
Addis Ababa farmers. All else equal, secondary town farmers use
The F-statistic is significantly different from zero, large in magni-
on average between 10 and 16 kg per hectare lower quantity of
tude, and is therefore well above the critical values suggested by
improved teff seeds. We also find that chemical fertilizer use is sig-
Stock and Yogo (2005). This suggests that the natural path variable
nificantly lower for secondary town farmers. These farmers use
is sufficiently strongly correlated to the endogenous transportation
between 82 and 125 kg per hectare less fertilizer than Addis Ababa
cost measure to make it a valid instrument.
farmers. Hence, these secondary town farmers are much less likely
to adopt modern inputs than primate city farmers. The distance to
6.2. Regression results the city is strongly negatively correlated to fertilizer and modern
seed usage for both Addis Ababa and secondary town farmers.
The empirical effect of cities and distance is evaluated through a The last two columns of Table 5 report the estimation results for
multi-variate regression analysis by using the models explained in yields (kg per hectare). In the reduced form specification, yields are
the previous section. In the first specification (column S1 in – on average – notably lower for secondary town farmers (260 kg
Table 5), we estimate a reduced form of Eq. (6) with only zone per hectare). This effect is quite large as the median yield in our
effects as control variables. In the second specification (column dataset is 1,179 kg per hectare (see Table 2). This finding is con-
S2), we control for household characteristics and prices for land firmed by the results of the second model specification where we
intensification measures. The first and second panel report the control for farm characteristics and prices, or when estimating
results from the OLS estimation and the 2SLS estimation (where the models with 2SLS to control for endogenous location of roads.
transport costs are instrumented), respectively. To improve read- Moreover, the distance from the farm to the city where teff is sold
ability of the results, we only report the regression coefficients of is an important determinant of productivity. An increase in trans-
the secondary town dummy and transport cost, the full regression portation cost of 10 ETB per quintal (which corresponds to an
table with coefficients can be found in an online appendix. increase of 50 km – see Fig. A1 in Appendix II) reduces yields –
The second and third columns of Table 5 report the regression all else equal – by almost 116 kg per hectare.
results of Eq. (6) for the teff output price (natural logarithms, with The results from the comparison of secondary town farmers
the coefficients multiplied by 1,000 to ease interpretation). Both with Addis Ababa farmers along several land intensification mea-
the reduced and full model in Table 5 indicate that teff prices are sures provide evidence that these outcomes, on average, are signif-
significantly lower for secondary farmers compared to Addis Ababa icantly lower for secondary town farmers. We find that the teff
farmers. The price that secondary town farmers receive for their prices that farmers in the rural hinterlands of secondary towns
teff in the secondary city is estimated to be, on average, 5 to 10 per- receive are significantly lower, and these farmers are also less
cent lower compared to Addis Ababa farmers. The distance to the likely to apply modern inputs in their production of teff. As a con-
Table 5
Transportation cost, cities, and teff intensification.
Price of teff Fertilizer (kg/ha) Improved Seed (kg/ha) Yield (kg/ha)

(ln of ETB/qnt)
S1 S2 S1 S2 S1 S2 S1 S2
OLS regression
Transportation Cost to the city (ETB/quintal) 3.43*** 2.71*** 3.66*** 2.50*** 0.39*** 0.41*** 15.64*** 10.32***
(0.64) (0.72) (0.63) (0.49) (0.11) (0.11) (1.66) (1.74)
Secondary Town (yes = 1) 65.10*** 54.46** 121.39*** 82.18*** 10.22*** 12.18*** 261.94*** 124.91*
(22.30) (20.61) (22.44) (14.91) (2.92) (3.24) (94.07) (69.82)
Constant 7,095.02*** 6,827.33*** 437.22*** 1,240.83 34.52*** 248.26 2,347.29*** 24,764.41***
(44.34) (154.57) (27.91) (1,793.85) (7.33) (237.99) (92.46) (7,055.98)
Prices No No Yes Yes Yes Yes Yes Yes
Zone fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
R-squared 0.257 0.307 0.280 0.477 0.210 0.328 0.175 0.317
2SLS regression
Transportation Cost to the city (ETB/quintal) 5.06*** 5.11*** 4.28*** 3.81*** 0.63*** 0.73*** 16.27*** 11.56***
(0.99) (1.02) (0.76) (0.56) (0.18) (0.20) (3.57) (2.98)
Secondary Town (yes = 1) 76.84*** 76.08*** 125.85*** 95.98*** 11.96*** 15.59*** 266.49*** 138.05**
(19.69) (18.03) (25.33) (15.92) (3.33) (4.04) (98.96) (69.68)
Constant 7,197.55*** 7,096.14*** 476.17*** –232.57 49.68*** 0.97 2,387.05*** –23,804.44***
(63.85) (161.13) (61.13) (1,787.86) (11.33) (268.57) (231.59) (7,013.53)
Prices No No Yes Yes Yes Yes Yes Yes
Zone fixed effects Yes Yes Yes Yes Yes Yes Yes Yes
R-squared 0.244 0.284 0.275 0.460 0.183 0.291 0.174 0.316
Durbin (score) chi2 8.23*** 17.73*** 3.19* 14.55*** 16.22*** 25.56*** 0.13 0.46
Wu-Hausman F test 8.25*** 17.70*** 3.17* 14.35*** 16.46*** 25.61*** 0.13 0.44
Robust regression 3.12* 7.50*** 0.53 6.72** 5.93** 5.62** 0.04 0.24
Note: number of observations in each model is 720. For each variable, S1 refers to the first specification of the reduced form model and S2 refers to the second specification of
the model with controls and prices (for the input regressions). The first panel reports the OLS estimates, while the second column reports the Instrumental Variable regression
(‘2SLS’) where transportation cost is instrumented by the natural path measure. The ‘Durbin’, ‘Wu-Hausman’ and ‘Robust regression’ test for endogeneity of transportation
cost are have H0: transportation cost is exogenous. Standard errors are clustered at village level and reported in parentheses below the coefficient: ***p < 0.01, **p < 0.05, *p <
0.1. The coefficients of the price regressions are multiplied by 1,000 to ease interpretation.
Table 6
Robustness checks.
Price of teff (ln of ETB/qnt) Improved Seed (kg/ha)

Robustness test 1 Robustness test 2 Robustness test 1 Robustness test 2
Coefficient s.e. Coefficient s.e. Coefficient s.e. Coefficient s.e.
Transportation Cost to the city (ETB/quintal) 4.21*** (1.35) 5.80*** (0.88) 0.83*** (0.20) 0.73*** (0.18)
Secondary Town (yes = 1) 66.78*** (20.12) 37.22* (21.05) 16.23*** (4.16)
Bahir Dar (yes = 1) 15.04** (5.95)
Nazareth (yes = 1) 16.23*** (4.50)
Travel time to Nazareth (hours) 0.83 (2.52)
Travel time to woreda administrative center (hours) 1.05 (0.88)
Travel time to cooperative (hours) 0.26 (0.99)
Magna (% of teff production) 9.19 (36.56) 6.38 (36.48)
Mixed (% of teff production) 91.23** (41.74) 70.41* (39.23)
Red (% of teff production) 54.82 (34.96) 21.48 (33.80)
High quality origin (yes = 1) 80.34*** (30.06)
Constant 7,009.65*** (162.64) 7,262.22 ***
(87.41) 9.74 (256.57) 35.50 (244.07)
Prices No No Yes Yes
Controls Yes Yes Yes Yes
R-squared 0.316 0.283 0.288 0.297
Durbin (score) chi2 14.38*** 48.34*** 27.32*** 27.42***
Wu-Hausman F test 14.22*** 50.31*** 27.41*** 27.44***
Robust regression 4.77** 27.99*** 6.21** 7.53***
Note: number of observations in each model is 720. For each variable, the coefficient of the 2SLS regression of second model specification is reported. The ‘Durbin’, ‘Wu-
Hausman’ and ‘Robust regression’ test for endogeneity of transportation cost are have H0: transportation cost is exogenous. Standard errors (s.e.) are clustered at village level
and reported in parentheses next to the coefficient: ***p < 0.01, **p < 0.05, *p < 0.1. The coefficients of the price regressions are multiplied by 1,000 to ease interpretation.
sequence, we observe lower yields in the hinterlands of secondary put prices in teff production, showing the robustness of our results
towns compared to those in the hinterland of Addis Ababa. We find to different specifications.
these effects in non-parametric and parsimonious regressions. The
effect of the secondary town dummy remains significantly nega- 6.3. Robustness test
tive in the IV specification (where transportation costs are instru-
mented with the natural path) or in the second specification We run two sets of additional regressions to assure further
where we control for household characteristics and input and out- robustness of our findings (Table 6). In both sets of robustness
tests, we report the 2SLS estimation of the most inclusive model, yields compared to their counterparts in the rural hinterland of
but again only report the coefficients of the main variables of inter- Addis Ababa. Our results therefore show that the location of farm-
est.21 First, although we control for the effect of teff quality on ers with respect to cities and the type of cities have strong effects
potential selection into a teff outlet market, teff quality might also on farmers’ intensification decisions in staple crop production.
have a direct effect on the teff price that farmers receive in the out- Our findings have potentially important implications for a
put market. In a first robustness test of the price formation model, broader welfare perspective. Our conceptual analysis combined
we add the shares of own production that farmers allocate to white with the empirical results suggests that there may be a trade-off
(default), magna, mixed or red teff varieties. We see that prices are in terms of the impact of the nature of urbanization (one primate
lower for lower quality teff varieties, but only significantly so for city versus multiple secondary towns) on agricultural develop-
mixed teff. Second, we add a dummy that equals one if the producing ment. The first effect is the positive influence from secondary
woreda is considered as a high quality teff origin. In both robustness towns. If secondary towns are absent in the rural economic space,
regressions, we observe that – even if we control for quality indica- and there is only the primate city where all urban consumers are
tors – farmers suppling to secondary towns receive significant lower located, our model predicts that at locations relatively remote from
prices, confirming the results of Table 5. the primate city, it becomes unprofitable to produce agricultural
Second, while modern seeds are distributed through farm coop- outputs for the urban market. Hence, these farmers are excluded
eratives in Ethiopia, the location of farmers with respect to the from the central market in the primate city and are most likely
agronomic research center in Debre Zeyt (where modern seed vari- to remain subsistence-oriented farmers. In the case where the
eties of teff are being tested) will potentially impact the uptake of urban population is not concentrated in one primate city but par-
modern inputs, ceteris paribus the effect of urban distance. There- tially distributed in secondary towns, the farmers who were ini-
fore, instead of including the secondary town dummy in Eq. (6), we tially located too far from the primate city to produce for its
first include dummies for each output market separately in the market are now influenced by the urban demand in the secondary
modern seed regression, with Addis Ababa the default. We see that towns. As a consequence, these farmers will start producing for
both farmers supplying Bahir Dar and Nazereth use significant less these urban markets and become responsive to market signals
modern seeds compared to Addis Ababa farmers, and while the dif- from them. Moreover, improved access to modern inputs could
ference is greater for Nazareth farmers, the difference is not signif- allow them to intensify their agricultural production.23 Hence,
icant (chi squared value of 0.03) from the uptake of Bahir Dar these non-linearities indicate that there is increased economic activ-
farmers. As a second robustness test, we include the travel time ity in the rural hinterlands because of the influence of secondary
to the research center in Debre Zeyt (close to Nazareth)22, as well towns.
as the travel time from the farm to the closest administrative center The second effect that we empirically observe in Ethiopia is that
and cooperative in the regressions, to control for potential additional the impact of the secondary towns is smaller than that of the larger
proximity effects (next to transport cost to the urban market). Sur- capital city. Hence, while more farmers may benefit from urban
prisingly, these (travel) distance indicators do not significantly affect spillover effects on agricultural prices and access to modern inputs
seed uptake. Hence, we find that farmers located around Nazareth do due to their proximity to secondary towns, the size of the benefits
not seem to benefit from their close proximity to the research center they realize may be smaller.
(at least in terms of modern seed uptake). The trade-off (or the net effect) depends importantly on the rela-
tionship between the size of the cities and the variables that matter
for farmers, such as agricultural prices and input markets. In terms of
7. Conclusions our conceptual model, the trade-off relates to the maximum value of
the I-functions in the theoretical models (see Figs. 2 and 3). The
Given rapid urbanization in developing countries, there is an model that we have developed provides a framework to identify
increasing interest in understanding the impact of the nature of these welfare benefits and costs, and can thus guide us to calculate
urbanization on the economies of these countries. Secondary and measure these welfare trade-offs in future research, based on
towns have been shown to lead to more inclusive growth and pov- the type of empirical estimates that we have provided in this paper.
erty reduction compared to primate cities. This is because rural
migrants are more likely to participate in the non-farm sector of
Conflict of interest
secondary towns (Christiaensen, De Weerdt, & Kanbur, 2016).
However, less is known about how urbanization patterns affect
None.
agricultural production. In this study, we investigate this relation-
ship between agriculture and different sized cities. A theoretical
model shows that output prices and intensification decrease over Acknowledgements
distance (measured through transport costs) to a primate city,
but that the presence of a secondary town introduces non- The authors thank Christopher B. Barrett, Luc Christiaensen, Todd
linearities in the relationship between these outcomes and urban Benson, and an anonymous reviewer for their helpful comments
proximity. and suggestions to the paper. We acknowledge the support for this
When we compare the model prediction with empirical obser- research by the Ethiopia Strategy Support Program (ESSP). ESSP is
vations, we find evidence that secondary towns influence the vari- managed by the International Food Policy Research Institute
ation of teff prices when transportation costs to the primate city (IFPRI), is implemented in partnership with the Ethiopian Develop-
increase. The empirical section of this paper further tests how ment Research Institute (EDRI), and is financially supported by the
the size of the city and urban proximity affects agricultural inten- United States Agency for International Development (USAID), the
sification. When we compare intensification between farmers close Department for International Development (DFID) of the govern-
to secondary towns and farmers close to Addis Ababa, we find that ment of the United Kingdom and the European Union (EU). The
secondary town farmers use less modern inputs and achieve lower research presented here was conducted as part of the CGIAR
21 23
The detailed regression output can be found in the online appendix. However, farmers that are located close to the primate city will be worse off when
22
The results do not change if we use the geometric distance from the farm to the the population is scattered over several secondary towns, compared to the case when
research center. all consumers are concentrated in the primate city.
Research Program on Policies, Institutions, and Markets (PIM), animal cost’. This transport cost was calculated from the total cost
which is led by IFPRI. spent by the farmer to travel from the farm to the self-reported
place of sales for each teff transaction, divided by the total amount
of teff sold to get the per unit transport cost (ETB per quintal). This
Appendix
cost depends on the characteristics of teff transaction, and the self-
reported cost could potentially be related to these characteristics
Appendix I: Sampling
or other factors. To avoid any endogeneity issues, we regressed
the transportation cost (ETB per quintal) on the mode of trans-
We use data of the 2012 large-scale survey of teff producers
portation, the place of sales and travel time to the place of sales;
located in five production zones around Addis Ababa (West Goj-
and subsequently use the predicted values. As most households
jam, East Gojjam, West Showa, South West Showa and East Showa)
performed multiple transactions, this was estimated using a fixed
which have the largest teff commercial surplus in Ethiopia. The
effects model. For those households that did not report any trans-
survey of in total 1,200 farmers was set up to be representative
actions, we used the village level data to replace the missing
of all land under teff in those zones. Farmers were randomly
values.
selected from both the smallest and largest teff producing woredas
As the teff is shipped from a trader town to a city by motorized
in these zones. To achieve this, all woredas were ranked in terms of
trucks, we used the ArcGIS software to calculate the transport cost
cultivated area within a zone. Two woredas then were randomly
using the road network in Ethiopia. The network analysis toolset
selected from the top 50% as well as from the bottom 50% produc-
was used to calculate the ‘truck cost’ for the second trip. A vector
ing woredas. Within each of the 20 selected woredas, all kebeles
layer of the road network of Ethiopia was obtained from the World-
(villages) were ranked in terms of teff production, and two kebeles
Map database (Guan, Bol, Lewis, Bertrand, Berman, & Blossom,
were randomly selected from the top 50% of teff-producing kebeles
2012),26 where for each road segment the total length as well as
and one from the bottom 50%. Hence, a total of 60 villages was ran-
the road classification is known.27 To construct a transport cost mea-
domly selected. Within each village, a census was created that
sure, we need data on transportation cost for each segment, which is
listed all farmers based on area cultivated. From this list, 20 farm-
not available. We overcome this limitation by calculating a per dis-
ers were randomly selected to be interviewed. Of these 20, ten
tance (km) cost of shipping teff over the segment in the following
farmers were selected from the list of large production farmers
way. First, data on transportation costs incurred by teff traders on teff
(cultivating all together 50% of the area) and ten farmers from
transaction trips with trucks were collected during complementary
small production farmers (the other 50% of the area). Hence, a total
surveys to this survey. This transportation cost data is the total cost
of 1,200 farmers was surveyed.
that traders and truckers in different trader markets faced when ship-
ping teff to Addis Ababa using trucks. Moreover, during the commu-
Appendix II: Calculation of transportation cost nity questionnaire, respondents were asked how much they think it
would cost to ship teff from the village center to Addis Ababa.28
We use detailed information on farmers’ self-reported teff sale Using the ArcGIS network analysis tool, we reconstructed the
or barter transactions for the teff harvested between September routes that these truckers and traders used from the market place
2011 and August 2012 to identify the local trader market for each they serve to Addis Ababa, following the roads digitalized in the
teff selling farmer.24 Farmers were asked what they consider the Ethiopian road network. For each route, we measured the travel
most common place of sales during the last production season. distance using the ‘Nearest Facility’ toolbox, which allows to calcu-
25% of the farmers responded that they sold the teff in the village, late a transport cost per unit of distance for each route. These
and 75% responded that the teff was sold in a trader market where routes were then overlaid with the road network to assign a per
wholesalers are present. We know the name of the market (town) distance transport cost to each road segment. Because of the spare
where the teff was sold for the latter group of farmers. For the farm- road network in Ethiopia, many of these routes overlap, implying
ers who sold teff in the village, and for farmers without teff transac- that for each road segment we have different observations of trans-
tion information, we used the community questionnaire information port cost. The average cost over different routes was calculated and
to identify the most common trader town used by teff farmers in used for each road segment. Finally, using the ‘Origin Destination’
each village. From a list of all markets visited by farm households toolbox in ArcGIS we calculated a distance, travel time and trans-
within the village, we used different information (e.g. the size of portation cost matrix for each pair of trader market – end destina-
the market and the share of teff traders in each market) to identify tion routes (i.e. Addis Ababa or the secondary town).
the most common teff trader town for each village.25 During the sec- The natural path distance is calculated in line with Damania,
ond trip, the teff is shipped from the trader market to a large whole- Berg, Russ, Barra, Nash, and Ali (2016) and Faber (2014). Using
sale market. For ach farmer, we use the city that was identified in the Digital Elevation Model of Ethiopia, we first calculated a ras-
Section 4 to be the most likely end destination of teff. ter with slope gradients. Then, we constructed a walking path
To calculate the cost associated with the teff transaction trip, friction surface raster by calculating for each pixel the estimate
we need to reconstruct (i) the farm to ‘trader market’ trip and (ii) time to cross the pixel on foot. Following Damania et al. (2016),
the ‘trader market’ to ‘regional market’ trip and measure the asso- we used the hiking velocity function proposed by Tobler (1993)
ciated costs. The cost of the first trip was collected from the price
transaction level information. Farmers were asked to report the
total cost (ETB) they had spent on transport for each transaction 26
Map obtained from https://worldmap.harvard.edu/data/geonode:roads_jgy.
27
of teff output (qnt). As donkeys are the most common mode to Based on the road quality, the following travelling speeds are assigned to each
transport teff to the trader town, this can be considered as a ‘pack road class: Motorable tracks (35 km per hour), dry weather roads (45 km per hour),
gravel all weather roads (60 km per hour) and asphalt all weather road (70 km per
hour). This allows to calculate the travel time between the trader towns and each end
24
However, 145 farmers (i.e. 19 percent) did not record any teff transaction. For destination.
28
these farmers, we will use data that was collected during a village level community Clearly, the main limitation of this data is that it only considers transport to Addis
questionnaire as explained later. Ababa and no other cities. However, as the road network in Ethiopia is sparse, many of
25
For the farmers that sold teff in a trader market, this data can be compared with the routes from markets to Addis Ababa overlap with the routes to other cities. Only
the self-reported market. For the majority of the farmers, these markets are the same, for the roads around the more remotely located villages, there might be data missing
but some farmers are participating in other markets than the most common village to ship teff to another city. These missing values were replaced by the average cost
market. per road class per zone.
Brutzkus, E. (1975). Centralized versus decentralized pattern of urbanization in

developing countries: An attempt to elucidate a guideline principle. Economic
Development and Cultural Change, 23(4), 633–652.
Cali, M., & Menon, C. (2013). Does urbanization affect rural poverty? Evidence from
Indian districts. The World Bank Economic Review, 27(2), 171–201.
Chamberlin, J., & Jayne, T. S. (2013). Unpacking the meaning of ‘market access’:
Evidence from rural Kenya. World Development, 41, 245–264.
Christiaensen, L., De Weerdt, J., & Kanbur, R. (2016). Urbanization and poverty
reduction: The role of secondary towns in Tanzania (No. 18). Antwerp, Belgium:
Institute of Development Policy and Management (IOB), Universiteit
Antwerpen.
Christiaensen, L., Demery, L., & Kuhl, J. (2011). The (evolving) role of agriculture in
poverty reduction—An empirical perspective. Journal of Development Economics,
96(2), 239–254.
Christiaensen, L., & Todo, Y. (2014). Poverty reduction during the rural–urban
transformation–the role of the missing middle. World Development, 63, 43–58.
Christiaensen, L., Weerdt, J., & Todo, Y. (2013). Urbanization and poverty reduction:
The role of rural diversification and secondary towns. Agricultural Economics, 44
(4–5), 435–447.
CSA (Central Statistical Agency), (2012). Agricultural sample survey: Area and
production of major crops, meher season (Vol. I) Addis Ababa, Ethiopia: Central
Statistical Agency.
CSA (Central Statistical Agency), (2013b). Population projection of Ethiopia for all
regions at Wereda level from 2014–2017. Addis Ababa, Ethiopia: Central
Fig. A1. Comparison of distance measures, i.e. transportation cost in ETB per Statistical Agency.
quintal (full line) or natural path cost in hours (dashed line). CSA (Central Statistical Agency), (2013a). Agricultural sample survey: Area and
production of major crops, meher season (Vol. I) Addis Ababa, Ethiopia: Central
Statistical Agency.
Damania, R., Berg, C., Russ, J., Barra, A. F., Nash, J., & Ali, R. (2016). Agricultural
to calculate the hiking velocity (V in km per hour) based on the technology choice and transport. American Journal of Agricultural Economics.
slope (S in gradients) of the terrain: aav073.
Djurfeldt, A. A. (2015). Urbanization and linkages to smallholder farming in sub-
V ¼ 6 e3:5jSþ0:05j Saharan Africa: Implications for food security. Global Food Security, 4, 1–7.
Dorosh, P., & Thurlow, J. (2013). Agriculture and small towns in Africa. Agricultural
A new raster file is created which calculates for every pixel the Economics, 44(4–5), 449–459.
Dorosh, P., & Thurlow, J. (2014). Can cities or towns drive African development?
‘cost’ (i.e. time) it takes to cross the cross the pixel following the Economy wide analysis for Ethiopia and Uganda. World Development, 63,
velocity function of Tobler (1993). Then, for each city, we calcu- 113–123.
lated the accumulated cost for each pixel to walk on foot to the city Faber, B. (2014). Trade integration, market size, and industrialization: Evidence
from China’s national trunk highway system. The Review of Economic Studies.
using the ‘Cost Distance’ tool in ArcGIS. This approach can be rdu010.
extended similar to Schmidt and Kedir (2009) to take into account Ferré, C., Ferreira, F. H., & Lanjouw, P. (2012). Is there a metropolitan bias? The
that it is impossible or more costly to cross rivers and lakes. These relationship between poverty and city size in a selection of developing
countries. The World Bank Economic Review, 26(3), 351–382.
cost layers were then overlaid with the location of farmers, so that Gibson, J., Datt, G., Murgai, M., & Ravaillion, M. (2017). For india’s rural poor,
for each farmer (pixel) we know the walking distance to each city. growing towns matter more than growing cities. World Development, 98,
Additionally, we computed least costly path using the ‘Cost Path’ 413–429.
Gollin, D., Jedwab, R., & Vollrath, D. (2016). Urbanization with and without
tool in ArcGIS, which visually shows the least costly walking route
Industrialization. Journal of Economic Growth, 21(1), 35–70.
for each farmer (which can then be compared with the actual road Gollin, D., Parente, S. L., & Rogerson, R. (2002). The role of agriculture in
placement). development. The American Economic Review, 92(2), 160–164.
Fig. A.1 below compares the different measure of distance. The Gollin, D., & Rogerson, R. (2014). Productivity, transport costs and subsistence
agriculture. Journal of Development Economics, 107, 38–48.
transportation cost (ETB per quintal) measured as the combination Guan, W. W., Bol, P. K., Lewis, B. G., Bertrand, M., Berman, M. L., & Blossom, J. C.
of the ‘pack animal’ and ‘truck’ cost, and the natural path distance (2012). WorldMap–a geospatial framework for collaborative research. Annals of
(hours) are plotted against the physical distance from the farm to GIS, 18(2), 121–134.
Henderson, J. V. (2010). Cities and development. Journal of Regional Science, 50(1),
the end destination of teff. We see that our measure of transporta- 515–540.
tion cost and natural path have a similar correlation with actual Henderson, J.V., Storeygard, A. and Roberts, M. 2013. Is urbanization in Sub-Saharan
distance. Africa different? World Bank Policy Research Working Paper, 6481.
Henderson, J.V., Storeygard, A. and Deichmann, U. 2014. 50 years of urbanization in
Africa: Examining the role of climate change. World Bank Policy Research
Appendix B. Supplementary data Working Paper, 6925.
Henderson, J. V., & Wang, H. G. (2005). Aspects of the rural-urban transformation of
countries. Journal of Economic Geography, 5(1), 23–42.
Supplementary data associated with this article can be found, in Ingelaere, B., Christiaensen, L., De Weerdt, J., & Kanbur, R. (2018). Why secondary
the online version, at https://doi.org/10.1016/j.worlddev.2018.03.006. towns can be important for poverty reduction – A migrant perspective. World
Development, 105, 273–282.
Jacoby, H. G., & Minten, B. (2009). On measuring the benefits of lower transport
References costs. Journal of Development Economics, 89(1), 28–38.
Jedwab, R., & Moradi, A. (2016). The permanent effects of transportation revolutions
in poor countries: Evidence from Africa. Review of Economics and Statistics, 98(2),
Bachewe, F., Berhane, G., Minten, B., Taffesse, A.S. (2018). Agricultural
268–284.
transformation in Africa? Assessing the Evidence in Ethiopia. World
Josephson, A. L., Ricker-Gilbert, J., & Florax, R. J. (2014). How does population
Development, forthcoming.
density influence agricultural intensification and productivity? Evidence from
Benziger, V. (1996). Urban access and rural productivity growth in post-Mao China.
Ethiopia. Food Policy, 48, 142–152.
Economic Development and Cultural Change, 44(3), 539–570.
Kedir, M., Schmidt, E. and Tilahun, H. 2015. Production patterns and fertilizer use.
Berdegué, J. A., Carriazo, F., Jara, B., Modrego, F., & Soloaga, I. (2015). Cities,
Mimeo.
territories, and inclusive growth: Unraveling urban–rural linkages in Chile,
Minten, B., Koru, B., & Stifel, D. (2013). The last mile(s) in modern input distribution:
Colombia, and Mexico. World Development, 73, 56–71.
Pricing, profitability, and adoption. Agricultural Economics, 44(6), 629–646.
Berhane, G., Paulos, Z. and Tafere, K. 2011. Foodgrain consumption and calorie
Minten, B., & Kyle, S. (1999). The effect of distance and road quality on food
intake patterns in Ethiopia. ESSP II Working Paper 23. International Food Policy
collection, marketing margins, and traders’ wages: Evidence from the former
Research Institute, Addis Ababa, Ethiopia.
Zaire. Journal of Development Economics, 60(2), 467–495.
Bloom, D. E., Canning, D., & Fink, G. (2008). Urbanization and the wealth of nations.
Minten, B., Tamru, S., Engida, E., & Kuma, T. (2015). Transforming staple food value
Science, 319(5864), 772–775.
chains in Africa: The case of teff in Ethiopia. The Journal of Development Studies,
Brückner, M. (2012). Economic growth, size of the agricultural sector, and
52(5), 627–645.
urbanization in Africa. Journal of Urban Economics, 71(1), 26–36.
Minten, B., Tamru, S., Engida, E., & Kuma, T. (2016). Feeding Africa’s cities: The case Econometric Models: Essays in Honor of Thomas Rothenberg (pp. 80–108).
of the supply chain of teff to Addis Ababa. Economic Development and Cultural Cambridge, UK: Cambridge University Press.
Change, 64(2), 265–297. Storeygard, A. (2016). Farther on down the road: Transport costs, trade and urban
Rashid, S., Tefera, N., Minot, N., & Ayele, G. (2013). Can modern input use be growth in sub-Saharan Africa. The Review of Economic Studies. rdw020.
promoted without subsidies? An analysis of fertilizer in Ethiopia. Agricultural Tobler, W. (1993). Three presentations on geographical analysis and modeling: Non-
Economics, 44(6), 595–611. isotropic geographical modeling speculations on the geometry of geography global
Ravallion, M., Chen, S., & Sangraula, P. (2007). New evidence on the urbanization of spatial analysis. California, USA: National Center for Geographic Information and
global poverty. Population and Development Review, 33(4), 667–701. Analysis.
Reardon, T. (2016). Growing food for growing cities: Transforming food systems in an Tschirley, D., Haggblade, S., & Reardon, T. (2013). Africa’s emerging food system
urbanizing world. Chicago, USA: The Chicago Council on Global Affairs. transformation. White Paper 2013 Series. Global Center for Food System
Reardon, T., & Timmer, C. P. (2014). Five inter-linked transformations in the Asian Innovation: Michigan State University, East Lansing, Michigan, USA.
agrifood economy: Food security implications. Global Food Security, 32, United Nations (UN) Population Division (2014). World urbanization prospects: The
108–117. 2014 revision. United Nations, New York, USA: Population Division.
Rondinelli, D. A. (1983). Towns and small cities in developing countries. Vandercasteelen, J., Dereje, M., Minten, B., & Taffesse, A. S. (2018a). Labour,
Geographical Review, 379–395. profitability and gender impacts of adopting row planting in Ethiopia. European
Schmidt, E., & Kedir, M. (2009). Urbanization and spatial connectivity in Ethiopia: Review of Agricultural Economics. https://doi.org/10.1093/erae/jby001.
Urban growth analysis using GIS. ESSP2 Discussion Paper 003. Addis Ababa, Vandercasteelen, J., Tamru, S., Minten, B., & Swinnen, J. F. (2018b). Cities and
Ethiopia: International Food Policy Research Institute (IFPRI). agricultural transformation in Africa: Evidence from Ethiopia. World
Stifel, D., & Minten, B. (2008). Isolation and agricultural productivity. Agricultural Development, 105, 383–399.
Economics, 39(1), 1–15. World Bank (2015). Ethiopia urbanization review: Urban institutions for a middle-
Stifel, D., & Minten, B. (2017). Market access, welfare and nutrition: Evidence from income Ethiopia. Washington, D.C., USA: The World Bank.
Ethiopia. World Development, 90, 229–241.
Stock, J. H., & Yogo, M. (2005). Testing for weak instruments in linear IV regression.
In D. W. K. Andrews & J. H. Stock (Eds.), Identification and Inference for

Are Rural Youth in Ethiopia Abandoning Agriculture

Uploaded by

Document Informationclick to expand document information

Copyright:

Available Formats

Are Rural Youth in Ethiopia Abandoning Agriculture

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

Are Rural Youth in Ethiopia Abandoning Agriculture

Uploaded by

Copyright:

Available Formats

World Development 106 (2018) 393–406

Contents lists available at ScienceDirect

1. Introduction In Africa, the share of the population that is urban, at 40 per-

Fig. 1. Impact of proximity to City 1 on I (prices and modern input use).

Variable Unit Data source Mean Median SD

5. Simulation results and observed teff prices

The theoretical model shows that if secondary towns influence

Price of teff Fertilizer (kg/ha) Improved Seed (kg/ha) Yield (kg/ha)

Price of teff (ln of ETB/qnt) Improved Seed (kg/ha)

Brutzkus, E. (1975). Centralized versus decentralized pattern of urbanization in

You might also like