LOps 3&(Sso
POLICY RESEARCH WORKING         PAPER        2845
The Spatial Division of Labor in Nepal
Marcel Fafchamps
Forhad Shilpi
The World Bank
Development Research Group
Rural Development
May 2002



POLICY RESEARCH WORKING PAIPER 2845
Abstract
Fafchamps and Shilpi examine how economic activity            crops are more important at intermediate distances.
and market participation are distributed across space.         Isolated villages revert to self-subsistence. The findings of
Applying a nonparanmetric von Thinen model to                 the study are consistent with the von Thunen model of
Nepalese data, the authors uncover a strong spatial            concentric specialization, corrected to account for city
division of labor. Nonfarm employment is concentrated         size. Spatial division of labor is closely related to factor
in and around cities, while agricultural wage employmenit      endowments and household characteristics, especially at
dominates villages located furtlher away. Vegetables are       the local level.
produced near urban centers. Paddy and commercial
This paper-a product of Rural Development, Developi-nent Research Group-is part of a larger effort in the group to
understand the importance of spatial factors in rural development. Copies of the paper are available free from the World
Bank, 1818 H Street NW, Washington, DC 20433. Please contact Forhad Shilpi, room MC3-536, telephone 202-458-
7476, fax 202-522-1151, email address fshilpti@worldbank.org. Policy Research Working Papers are also posted on the
Web at http://econ.worldbank.org. The authors may be contacted at marcel.fafcliamps@)economics.ox.ac.uk or
fshilpi@?worldbank.org. May 2002. (62 pages)
The l'olicy Research Working Paper Series dissenimates the findings of wvork in progress to encourage the exchange of ideas aboiut
developmzent issues. An objective of the semes is to get the findings out quzickly, even if the presentations are less than fully polished. The
papers carry the names of the authors and shotuld be cited aicordingly. Th2e findings, interpretations, and conclusiois expressed in this
paper are entirely those of the authors. They do iiot necessarily represent tl7e viewo of the World Bank, its Lxecitive Directors, or the
countries they represent.
Produccd by the Research Advisory Staff



The Spatial Division of Labor in Nepal'
Marcel Fafchamps
University of Oxford
Forhad Shilpi
The World Bank
Abstract
JEL classification: RI 1, 018
Keywords: economic geography, South Asia, specialization, roads
The Policy Research Working Paper Series disseminates the findings of work in
progress to encourage the exchange of ideas about development issues. An objective
of the series is to get the findings out quickly, even if the presentations are less than
fully polished. The papers carry the names of the authors and should be cited
accordingly. The findings, interpretations, and conclusions expressed in this paper
are entirely those of the authors. They do not necessarily represent the view of the
World Bank, its Executive Directors, or the countries they represent.
We would like to thank Gershon Feder for his constant moral and intellectual support. We acknowledge
generous support from the former Chief Economist of South Asia Region John Williamson. Road distances
and population densities were calculated with the help of Uwe Deichmann, Jyotsna Puri and Andrew
Nelson at the GIS Lab of the Devetopment Economics Research Group (DECRG) of the World Bank. We
would like to akowledge Giovania Prennushi and Salman Zaidi for NLSS data and complementary
documentation and Lawrence Zhangj and Misuzu Otsuka for excellent research assistance. We also received
useful comments on an earlier draft from seminar participants at World Bank. Financial support from the
Development Economics Research Grout (DECRG) is also gratefully acknowledged.






1. Introduction
After decades of neglect, the economics profession has rekindled its interest in geographical
phenomena. Much recent work has been devoted to agglomeration effects across regions and
countries (e.g. Krugman 1991b, Fujita, Krugman and Venables 1999a, Ciccone and Hall 1996).
Apart from research focusing on cities (e.g. Henderson 1974, Lamnpard 1968, Abdel-Rahman 1994,
Rauch 1993, Fujita, Krugman and Mori 1999b, Arthur 1988, Glaeser, Kallal, Sheinkman and
Shleifer 1992, Ellison and Glacser 1997, Glaeser, Scheinkman and Shleifer 1995), most empirical
work has remained at a fairly aggregate geographical level (e.g., Ciccone and Hall (1996), Radelet
and Sachs (1998), Hummels (1995), Hall and Jones (1996) on countries; Desmet (1998) on
European regions; Barro and Sala-I-Martin (1991) on U.S. states). Little attention has been
devoted to more disaggregatel geographical units that have a predominantly rural character,
such as villages or counties (see, however, Ciccone and Hall (1996), Ciccone (1997), Fafchamps
and Helms (1996), Desmet and Fafchamps (2000)). This is somewhat surprising given that
much of traditional economic geography focuses on spatially disaggregated phenomena such as
the distribution of cities over space and their relationship with surrounding rural areas (e.g. von
Thunen 1966, Isard 1956, Christaller 1966, Losch 1954, Dicken and LLoyd 1990, Jacobs 1969).
There is also little recent work by economists on the spatial distribution of economic activity
across space in rural and urban areas of the Third World, in spite of a long tradition of research
on regional issues among development economists. This is not to say that there has been no
work on spatial issues. The contrast between cities and countryside has long attracted the
attention of development economists, to the point that this contrast has become a fundamental
organizing concept of all development theory (e.g. Lewis 1954, Harris and Todaro 1970). There
has also been a lot of work on work migration, non-farm production in rural areas (e.g., survey
by Reardon (1997); for Nepal, see Seddon, Adhikari and Gurung (1999)), and on the spatial
2



integration of agricultural markets (e.g. Takayamna and Judge 1971, Ravallion 1986, Dercon 1995,
Timmer 1986, Baulch 1997, Fafchamps 1996). But this work remains fragmentary in the sense
that it does not examine, in a comprehensive manner, how rural areas interact with each other
and with cities as a function of their geographical proximity to cities.
The purpose of this paper is to begin filling this lacuna by examining how Nepalese households
fit into the local economy as a function of their proximity to urban centers. Nepal is a particularly
suitable place to study spatial specialization given the extreme diversity of the country in terms
of accessibility. At one end of the spectrum, Terai villages are a few hours away by truck from
many towns and cities, while Nepalese villages tucked in the Himalaya are among the most
difficult to access anywhere on earth. As already argued by Jacoby (2000), Nepal is the perfect
place to examine the effect of geographical isolation on economic activity.
Using household data from the Living Standard Measurement Survey (LSMS) of 1995/6, we
investigate geographical patterns of agricultural production, agricultural sales and purchases,
and non-farm work. To control for differences in road quality, travel time is used as a measure
of distance instead of mileage. City population and travel time are instrumented to control for
possible endogeneity.
Our contribution is threefold. First, by using several different measures of economic activity
and market participation, we provide a detailed and compresensive picture of the spatial division
of labor in a poor country. Second, thanks to non-parametric methods, we obtain precise
estimates of the distance at which various activities dominate. As a result, we can characterize
the various forms that market specialization takes depending on location. Third, we show that
spatial effects are large and that city size matters.
Regarding spatial specialization, our results largely confirm the von Thilnen hypothesis of
concentric circles around cities that vary according to transport costs (e.g. Henderson 1988,
3



Dicken and LLoyd 1990, Abdel-Rahman 1988, Fujita 1988). Nepalese households living near
markets and cities are more likely to engage in non-farm (wage) work rather than prodluce agri-
cultural goods for sale. Fears that cities 'steal' non-farm jobs from the surrounding countryside
do not appear justified in the case of Nepal. If anything, it is proximity to markets and towns
that fosters non-farm activity, especially wage non-farm employment.
As one moves away from markets and towns, the emphasis shifts progressively from non-farm
activity to agricultural production for sale. As anticipated, vegetables tend to be produced close
to cities because of their perishability while commercial production of storable crops such as rice
and pulses takes place furlher away. Minten and Kyle (1999) obtain a similar result in Zaire.
The emphasis on commercial farming also raises hired labor use. Households located close to
towns and markets rely primarily on the market for their consumption needs. Labor migrations
are also more frequent in these villages but the effect is not strong.
As one moves further away from towns and markets, households revert to self-subsistence for
crops but shift to livestoch production as a source of cash income. This is because animals and
animal products (e.g., ghEe) are more easily transported to distant markets. Isolated villages
do not interact with the market at all. These results confirm earlier work in Nepal by Jacoby
(2000) and Seddon et al. (1999).
Regarding the spatial range over which certain activities dominate, we find that non-farm
wage employment takes place primarily within four hours travel time from large towns. Within
villages, non-farm employrment is concentrated within one hour of the nearest market. Vegetable
production is important in the 1-3 hours range, while the production of paddy for sale dominates
the 3-5 hours range. Beyond 5 hours travel time from cities, households no longer sell crops and
rely on production to satisfy consumption needs. They also stop purchasing fertilizer. For those
located more than 5 hours from a market or city, cash income can still be obtained from livestock
4



but households located more than 10 hours from the market essentially revert to self-subsistence.
To our knowledge, this is the first time that a comprehensive quantification of these distance
ranges has been presented for a poor country.
Spatial effects are very strong and robust determinants of activity choice and market par-
ticipation. For instance, a household located right next to a rural market would allocate 40%
of its time to non-farm self-employment and 30% to non-farm wage employment compared to
0% for a household located one hour away from the market. Ignoring such spatial effects would
undoubtedly bias poverty estimates.
Spatial effects are also important for policy. Our results, for example, show that households
located more than 5 hours away from towns and market do not sell crops and do not consume
fertilizer. Any effort to promote agricultural innovation would need to take these effects into
account to be successful.
The paper is organized as follows. Concepts and theory are discussed in Section 1, together
with the econometric approach. The data and its main characteristics are presented in section 2.
We also describe how our various measures of geographical proximity are constructed. Section 3
tests the effect of distance to markets and cities. Section 4 investigates the role of factor endow-
ments and preferences in activity choice and market participation. Conclusions are summarized
at the end.
2. Theory and Concepts
German economist von ThtOnen was the first to hypothesize, as early as 1842, that economic
activity need not be spread evenly across space even if land is undifferentiated between loca-
tions. His basic idea was that rural areas surrounding cities specialize in different agricultural
products. He further argued that the product they specialize in depends on the cost of trans-
5



porting output to the marl:et. Locations close to the city specialize in high transportation cost
goods, such as milk and vegetables, while locations further away specialize in less perishable,
lower transport cost commodities such as cereals and pulses. Rural communities klcated too
far from cities to trade with them must turn to self-subsistence in both agricultural and non-
agricultural commodities Erugman (1991b), the latter being produced using small-scale, artisan
technology. This theory is usually represented graphically as concentric circles of specialization,
beyond which lies undifferentiated self-subsistence areas. Adding to this theory, Dore (1987) and
Jacobs (1984) argue that peri-urban agriculture also benefits from spillover effects in technol-
ogy and marketing, thereby suggesting something like concentric circles of technology usage as
well. Another addition is the realization that isolated communities can interact with the global
economy by sending workers away for extended periods of time, thereby triggering patterns of
long-term work migrations to cities or plantations.
Von Thilnen's theory was further refined by Losch (1954), Isard (1956) and Christaller (1966)
who studied the relationship between cities of various sizes and between them and surrounding
rural areas. They hypothesize the existence of hierarchies of cities performing different roles, each
with its own rural hinterland (see Lee (1993) and Fafchamps and Helms (1996) for illustrations in
Mexico and Guatemala). These refinements do not qualitatively affect the spatial specialization
idea, although they may alter the shape it takes (e.g., honeycomb instead of circle). They also
bring out the role that city size has on the width of concentric circles of rural specialization, larger
cities requiring more vegetable and food products than small cities. More recently, it has also
been shown that, with two or more immobile factors instead of one (e.g., land), more complex
patterns can be generated that include incomplete or partial specialization (e.g. ?, Venables and
Limao 1999).
The recent economic geography literature has revisited many of these themes (e.g. Henderson
6



1988, Krugman 1991a, Fujita et al. 1999a). Much attention has also been devoted to the inner
organization of cities themselves (e.g. Fujita 1988, Abdel-Rahman 1993, Abdel-Rahman 1994,
Henderson 1974). In this respect, the literature has emphasized one important organizing prin-
ciple, namely the tension between, on the one hand, agglomeration effects that incites firms to
cluster, and, on the other, housing prices and commuting costs that raise workers' wages. This
has, for instance, led to models of cities whereby activities that benefit from strong agglomera-
tion effects, such as financial services, take over city centers while workers live at the periphery
where housing costs are lower. Depending on travel costs for consumers, shopping districts locate
in city center or in residential areas (e.g. Abdel-Rahman 1988, Fujita 1988). In these models,
the size of cities depends critically on their ability to attract workers. This in turn depends on
the strength of agglomeration effects, which affects return to labor, and on commuting costs and
housing prices, which affect wages. Further refinements, such as pollution, congestion, and the
provision of intermediate goods can be added to generate different structures, e.g., industrial
basins surrounded by residential towns some distance away (Desmet and Fafchamps 2000).
All these theoretical predictions can be summarized as follows. In its simplest form, the
concentric circle theory predicts that what a community produces depends on distance from the
nearest town. Villages located nearby town centers are expected to produce perishable products
with a high transport cost, such as vegetables, while villages located further away are expected
to produce low transport cost commodities such as cereals and pulses. A variant suggested by
the urban economics literature has workers reside away from their work place. In this variant,
villages and neighborhoods located close enough to cities count a number of commuters, that
is, individuals who work in the city but return to the village (the suburb) at night. The income
they bring to their place of residence in turn generates suburban jobs in consumer services and
retail trade - what could be called the 'shopping mall syndrome'.
7



Theories of hierarchies of cities predict more complex concentric zones whose width and
spatial structure depend on their interaction with various cities of different sizes. In particular,
they predict that larger cities have a larger hinterland, and that rural dwellers may buy and
sell from different cities sirmultaneously. The presence of more than one immobile factor leads
to incomplete specialization, whereby villages produce a multiplicity of goods at the same time.
Finally, all theories predict that isolated locations must be self-sufficient in both agricultural
and non-agricultural products, except for long-term work migrations.
These predictions can be empirically investigated as follows. Consider a vector of measures
of economic specialization and market participation for individual households i. Let this vector
be denoted {1y } for z = 1, ..., Z, where yil is, say, vegetable production, y2 is non-farm self-
employment, y3 is rice purchases, etc. Further suppose that we have information on the distance
to the market center nearest to household i, denoted di. The von Thilnen hypothesis, in its
simplest form, predicts a relationship between y1 and d,. As is clear from the economic geography
literature, this relationship is expected be non-linear, with unknown inflection points. The
relationship can be written:
yi4  f (d) + ±ui                              (2.1)
where ui is an error term and f(di) is an unknown smooth function of distance to the nearest
market d,. A similar approach is adopted by Chomitz and Gray (1996) in their analysis of
land use in Belize. Estimating function f (.) non-parametrically provides a simple way of testing
various hypotheses about the effect of location on economic activity. In addition, inflection
points and zeros of the fitted f (.) function provide estimates of the width of various concentric
circles, hence providing useful information for policy makers.
Equation (2.1) presen1s a number of drawbacks, however. First, it does not control for town
8



size. Yet refinements of the von Thflnen hypothesis predict that the width of concentric circles
and the strength of the spatial division of labor depends critically on the magnitude of urban
demand and thus on town size. Second, the equation fails to account for the possible effect of
more distant cities and towns. As Isard (1956) has shown, the organization of economic activity
over space reverts around a hierarchy of cities and partially overlapping concentric 'circles' (or
rather, hexagons). Recent contributions also emphasize a possible overlap between the effect of
multiple cities (Fujita et al. 1999b). A better test of the theory should thus allow for town size
and the possible effect of multiple towns. Such a test can be constructed provided we have data
on urban population residing at various distances h from household i. Let this information be
denoted as {pi(h)}. An alternative, more general test of spatial division of labor can then be
constructed by estimating equations of the form:
H
Yi= f (d.) +   g(h)pi(h)dh + ui                     (2.2)
The presence of two unknown functions in equation (2.2), the fact that function g(.) is
multiplied by population pi(h), and the presence of censoring in the dependent variable make
estimation by conventional non-parametric techniques very difficult. To turn the above equation
into an estimable regression model, we choose instead to discretize both functions f (.) and g(.):
J         H
Yi =     rjD   + hP   +ui                         (2.3)
j=1       h=1
where £i is a dummy variable taking the value 1 if j-1 < di < j, and 0 otherwise. pFh is
the urban population residing within, say, h and h - 1 hours of travel from household i, i.e.,
pih == fh_ pi(s)ds. Parameter H is chosen large enough that proxmity effects die out, that
9



is, such that g(H) - 0. Estimnation efficiency can be improved by requiring that the estimated
rj and 'Yh parameters generate smooth approximations for functions f(.) and 9(.). One such
method is the so-called roughness penalty method suggested by Good and Gaskins (1971) and
Silverman (1982). In the case of ordinary least squares, the estimator is obtained by minimizing:
T        J         H           J-1
>JYi -E   iTDi -E ZYhr] + E>        A[(,rj+_- rj)-(j- j_    _)]2
i=1     j=1        h=1          j=2
H-1
+      A [(Yh.+l -Yh) - (-Yh -fh-1)]
h=2
where T is sample size and ), and A, are penalty parameters.' All reported standard errors are
robust (White) standard eriors corrected for clustering.2 The purpose of the rest of the paper
is to estimate the above model using data on Nepal.
In case yil is dichotomous so that OLS is inappropriate, the sum of squared residuals in
equation (??) is replaced by the required likelihood function. Penalty parameters Ar and A..
are adjusted accordingly. When the estimating function is a likelihood function (and provided
some other conditions are satisfied), Silverman (1984) has shown that the roughness penalty
approach yields a kernel estimator of f(.). When the dependent variable is censored, we use
OLS instead of tobit for two reasons. First, we are interested in how distance affects the
unconditional expectation cf yil, not how it affects yi' conditional on being uncensored. Second,
in the presence of heteroskedasticity and clustering, tobit is inconsistent and standard errors
cannot be corrected.
1These estimates of the r's and -y's can easily be obtained using the regular OLS command by adding J + H -4
artificial observations at the end of the sample such that dependent variable and regressors are zero, except for
Dn_T-    A_ Dn_T =-2A,., and D.T+l = A, for n = T + 1 to T + J -2; and Pn -T-1- A  pn-T =
and npT+ -A.for n = T +J - 1 to T + J + H -4.
2As is often the case in survey data, correcting for hetereskedasticity and clustering makes a big difference. To
check the possible impact of the roughness penalty correction on standard errors, we compared reported standard
errors with standard errors obtained by bootstrapping with 200 replications (with replacement). Given the large
sample size and low level of smoothing, bootstrapped confidence intervals are virtually indistinguishable from
OLS estimates.
10



3. The Data
Home to the Everest, Nepal is located nearly entirely at the foot of the Himalaya mountains. It
is composed of essentially four regions: the Himalayas themselves, which run along the entire
Northern boundary of the country and are very sparsely populated; the central valley around
Kathmandu; mountains (locally called 'hills' to contrast them with the Himalayas) that run
East-West parallel to the Himalayas; and the plain of Terai that borders India to the South and
has the best agricultural land (Government of Nepal 1995).
Nepal is largely rural, with 86% of its 21 million inhabitants living in villages or towns of
less than 10,000 people. Kathmandu, the capital city and largest urban center, has a population
of around half a million people.3. There are only 34 cities and towns of 10,000 inhabitants or
more, most of which can be found either in the central valley or in Terai. Given the mountainous
terrain, communications are generally more difficult within Northern Nepal itself than within
Terai. In fact, people living in the remote Northern part of Nepal must trek many hours by foot
or bullock cart before reaching the nearest road. Nepal thus offers a perfect testing ground to
examine the effect of isolation on spatial specialization and market participation.
The data we use comes from the Living Standard Measurement Survey (LSMS) of 1995/96.
The survey covers some 3373 urban and rural households spread among 274 villages or 'wards'
distributed over all regions of the country (Figure 1 - Map of Nepal with location of surveyed
villages). 28 of these wards are located in Kathmandu alone. As with other LSMS surveys,
data coverage is quite comprehensive. For our purpose, information is available on agricultural
production, cropping patterns, self-employment in non-farm activities, wage employment by
sector, sales and purchases of crops, and migrations. Unlike other LSMS surveys, the Nepal
survey also contains detailed information about distance and travel time to markets and towns.
3421,000 inhabitants in 1991.



Jacoby (2000) has used this information to show that land prices fall with distance from markets.
Table 1 summarizes the main characteristics of surveyed households. We see that households
are mostly nuclear, with a large age gap between the household head and his spouse. Households
own a couple heads of livesto(k and cultivate one hectare of land on average, a third of which is
irrigated. Educations levels are very low. Given that settlement is dispersed in much of Nepal,
the location of individual hovuseholds within the same ward varies a lot. The average household
is located two hours away from the nearest market. The median is lower, at just above one
hour. In the econometric analysis, the 20 distance dummies D: are constructed by dividing the
sample into five percent percentiles.
Table 2 summarizes the measures of economic activity and market participation that we were
able to construct on the basis of the available data. We use seven categories of variables. Data
on labor form the basis for our first set of specialization and market participation measures.
The share of farm work in total employment is taken as rough measure of specialization in
farming. Presumably non-farm work is highest either in isolated areas - where everything has
to be produced locally - ard at the proximity of urban centers - due to the joint effects of
commuting and the shopping mall syndrome mentioned in Section 1. In isolated areas where
the size of the market is small, we would expect that most non-farm production takes the form
of small enterprises - and hence of self-employment in non-farm activities. Where the market is
larger, e.g., in urban centers, larger enterprises could take advantage of returns to scale, hence
more emphasis on wage non-farm employment. By the same token, we expect administrative,
clerical, and professional jobs to be concentrated in towns where large firms and government
offices are concentrated.
Other patterns are also possible. For instance, it is conceivable that in isolated areas, most
non-farm production takes place within the household itself and is not counted as work (e.g.,
12



food preparation, fuel and water collection, child care, personal services). In this case, non-
farm work would not be recorded as such in self-subsistence regions (see Fafchamps and Helms
(1996) for a similar observation in Guatemala). Prosperous agricultural areas may also foster
the emergence of an active non-farm sector, especially in the production of non-tradables.
Table 2 shows the share of total household labor going to farm and non-farm work, both
for self-employment and wage work. Self-employment in farming is the dominant form of em-
ployment, a reflection of the predominantly rural nature of the sample. Wage employment
outside agriculture is the second most important category. Agricultural wage work and non-
farm self-employment each represent about 12 percent of all labor. The breakdown of non-farm
employment by sub-sector shows that trade and manufacturing (crafts) are the most impor-
tant self-employment sectors. Production dominates in wage employment, together with various
clerical, administrative, and professional jobs.
A measure of work-related short-term migrations is also given in the Table. It is constructed
as the number of household members who work outside the survey ward. This is a reasonable
approximation given that, to be counted as household member, a person must reside with
the household. We expect migration to be most prevalent among households who reside at
the vicinity of urban centers, although workers may also commute to neighboring agricultural
employment areas.
Other measures of specialization focus on farming for which the survey contains a lot of
information. Cultivated area and sale of livestock products capture the relative emphasis on
agriculture or livestock. Crop choices are measured by cropping patterns, that is, by the share
of land planted to different crops. Given the dependence of actual crop output on the weather,
cropping patterns are a better measure of intended output than output itself. By definition, they
are only available for farming households. Paddy, pulses, and vegetables are the main crops in
13



the wet season. During the dr.y season, rice is typically replaced by other cereals because they
are less dependent on water. Vfegetables are the second most important crop, foLLowed by pulses
and oiLseeds. Data on fertilizer use is meant to capture the relationship between location and
technology adoption.
Market participation for agricultural products is presented in the second half of the table.
We measure the extent to which surveyed households rely on the market to dispose of their
excess agricultural output or to meet their food needs. We see that crop commercialization
remains low and that most surveyed households consume most if not aLL of their output. Sales
are somewhat higher for crop;3 other than paddy and cereals. The situation is reversed in terms
of purchases: the average survey household purchases half of the rice and one third of cereals it
consumes. The bottom of the table further confirms that the median Nepalese household has
a large food deficit, particularly in rice and vegetables. Three quarters of them (two third of
all farmers) spend more on rice than the value of their paddy output; two third of them (55%
of aLL farmers) spend more on agricultural products than the value of their agricultural output.
Marketed surplus is producecL by a small group of farmers. Barrett and Dorosh (1996) describes
a similar situation in Madagascar. The geographical location of this marketed surplus is of
obvious interest to policy makers.
We complement these L'SMS data with information about urban population in Nepal (34
towns and cities).4 For our purpose, a town is defined as a settlement of more than 10,000
inhabitants. We first compate the distance between each surveyed ward and each of these
towns. Distances are normaLly taken along existing roads, except when roads do not exist, in
which case we calculate the shortest arc distance to the nearest road, and then the distance
41n earlier versions of this paper, we also included population in Indian towns located within 100 km of the
Nepalese border. Including them does not affect qualitative results, but their effect on spatial specialization in
Nepal is small. They are omitted Irom the paper because instrumenting urban population and distance for Indian
towns raises data and computation requirements beyond what we could haridle. This issue is left for further
research.
14



to various cities along the road.5 Distances are then converted into travel time using available
information about trucking and walking speeds along various types of roads in Nepal.6 Off the
road travel is assumed to take place by foot - a reasonable assumption for Nepal given the
nature of the terrain.
Available information on distance to towns is summarized in Table 3. The average distance
from surveyed wards to the nearest town is just under 4 hours, with large differences across
wards. Around 30% of surveyed wards are located either within towns or very close to towns.
Close to half the surveyed households live at most two hours travel away from a town or city;
the median distance is 2 hours and 12 minutes. The other tail of the distribution is very long,
reflecting the mountainous and isolated nature of much of Nepal. Twenty percent of surveyed
wards are located more than 7 hours travel from the nearest town; ten percent are more than
10 hours away. One surveyed household in 30 is 15 to 30 hours travel away from the nearest
town. This much variation should make it easier to identify the effect of distance on economic
activity.
Combining information on distance to towns with data on population in these towns, we
construct a measure of urban population at various time distances from each ward. These
measures, denoted pih for h = 1, ..., H hours, only vary from ward to ward. They are organized
as follows. Suppose that a ward i is 3 hours away from the nearest town, which has a population
5This is a very time consuming process that requires a combination of various techniques. e.g., visual in-
spection of maps, statistical information on road grades, calculation of arc distances, comparisons across various
measurements to identify shortest distances, etc. The assistance of Uwe Deichman and Jyotsna Puri (GIS lab,
Department of Research of the World Bank) was essential to the success of this operation.
6Travel speeds are calculated for various terrains and types of road. Assumed travel times are as follows, in
km/hour:
Highway   Provincial road  Secondary road  Off road
Terai              60           35             10           5
Siwalik            51          29.75           8.5         4.25
Middle mountain    42          24.5             7           3.5
High mountain      36           21              6           3
High Himalayas      30         17.5             5           2.5
These figures were obtained through discussion with various transportation experts and South Asia operations
staff at the World Bank. Travel on highways and provincial roads is assumed to take place by truck; travel on
secondary roads is assumed to be by cart.
15



of 30,000. The next nearest town is 7 hours away and has a population of 100,000. In this
case we have, for each household in the ward, {Pil; ...; PFl'} = 0;0; 30,000;0 0;0; 100,000;0,0,0.
Table 4 summarizes our constructed pih variables. The average surveyed ward has an urban
population of 75,000 inhabitants located within an hour travel time. The median, however, is
zero. Urban population first goes down with distance, reflecting the fact that some surveyed
wards are urban or peri-urban. It then increases steadily, as more and more towns fall within a
given travel time radius from the ward. We confine our analysis of urban population effects to
a 10 hour radius; beyond 10 hours travel time, urban proximity effects taper off. Population is
measured in millions.
Before proceeding with the analysis, we must correct for the possible endogeneity of urban
population and travel time. It is indeed conceivable that towns are larger whenever the sur-
rounding countryside prod aces a food surplus. Observing that wards located to large towns sell
more food would then be the result of reverse causation. The same reasoning applies to road
construction: public authorities might be more inclined to build roads to areas that produce
more agricultural surplus. Since travel is faster on roads, this would result in surplus areas
being closer to towns in terms of travel time. Similar concerns about reverse causation apply to
non-farm production, vegetable production, etc.
To address these conce.rns, we instrument city population and travel time as follows. Pre-
dicted urban population is obtained by regressing the log of actual population on physical char-
acteristics of the district in which the town is located: the log of its size in square kilometers;
the log of its arable land area; the log of the distance to the nearest river; its mean elevation;
the standard deviation of the elevation within the district; and a dummy if the district is the
mountainous part of the country. By limiting our instruments to physical features, we minimize
16



the risk of that instruments are themselves endogenous. 7 Size is included because larger dis-
trict can hold more people. Arable land area proxies for food production potential. Distance to
the nearest river proxies for accessibility, as roads often follow river valleys. Elevation controls
for climate - towns are less likely at higher elevations. The standard deviation of elevation is
a measure of roughness in the terrain; towns are more likely in flat districts. The mountain
dummy is included for the same reason. Regression results are presented in Appendix 1. The
R2 of the regression is 0.27. Regressors in general have the expected sign but are multicollinear
(mean VIF of 8.7), which explains why regressors are not individually significant. Since we are
only interested in predicting population, multicollinearity does not matter.
Travel time between a ward and a town is instrumented using foot travel time as well as
physical characteristics of the ward and town: size of the district; arable land; mean elevation;
distance from the nearest river (available only for the town); standard deviation of elevation;
and regional dummies (East-West). Foot travel time is computed using iso-elevation curves to
account for the mountainous nature of the terrain. The regression is estimated in log form.
Results are presented in Appendix 2. As expected foot distance is the major determinant of
travel time. The R2 of the regression is 0.84. Population variables ph used in the subsequent
analysis are constructed using predicted urban population and travel time instead of actual
values.8 As it turns out, results are virtually identical to those obtained with actual values,
suggesting that the potential endogeneity of town population and road construction are not a
source of bias in these data.
'We also experimented with a longer list of instruments, including area of irrigated land and the like. This
results in a better fit for town population but the rest of the analysis is unaffected. Additional regressors are
omitted in the analysis presented here to minimize the risk of endogeneity and overfitting.
8Standard errors are not corrected for the use of predicted regressors. Doing so would be extremely difficult
given the way in which regressors are constructed. The fact that instrumented and unininstrumented results are
very similar suggest that inference is very unlikely to be affected.
17



4. Econometric Results
We now turn to the econometric analysis. We only present selected Figures and Tables. We
begin with employment patterns. The dependent variable is the share of household labor going
in non-farm and farm activities (self-employment or wage work); it is our measure of economic
emphasis on non-farm and farm production. Since respondents are interviewed at their place
of residence and most cormnute to work by foot, we expect reported employment patterns to
reflect working conditions in the vicinity of the place of residence. We experiment with various
values of the penalty parameter A. Our best results (i.e., neither too smooth nor too rugged) are
those presented. Qualitative results are not very sensitive to the value of the penalty parameter.
Results are presented in graphical form to facilitate interpretation. All Figures report coefficient
estimates as well as the bootstrapped 95% confidence interval around them.9
The shape of function Jf(.) is shown in Figure 2 for each employment variable; the shape of
function g(.) is shown in Figure 3. Distance to towns varies across wards only; the distance to the
nearest local market varies across households within the same ward. Function f(.) thus controls
for within ward spatial variation. Figure 2 (and all subsequent figures relative to within-ward
spatial variation) uses distance percentiles, not actual distance, on the x axis.10 The effect of
actual distance is much steeper than shown. With this caveat in mind, the resemblance in the
shape of both functions is cunning. Non-farm wage employment decreases sharply with distance
from market and urban centers. Within a ward, it tapers off within one hour walking time from
the nearest market (Figure 2). Non-farm self-employment also decreases with distance from the
local market but it is not significantly affected by proximity to urban population (Figure 3).
9Surveyed wards represent only a small proportion of all wards and are located far apart. It is therefore
extremely unlikely that standard errors suffer from bias due to spatial autocorrelation. Correcting standard
errors using, for instance, the method proposed by Conley (1999) is not necessary. Reported standard errors do,
however, correct for cluster effects, that is, correlation in household residuals within wards.
'0Dummies used in estimation - and thus confidence intervals - are constructed using percentile intervals.
18



These results indicate that proximity to urban population raises the probability that a house-
hold is involved in non-farm wage employment. Wage employment outside agriculture thus ap-
pears to be primarily an urban phenomenon. Within a ward, households involved in non-farm
employment are located in or very near market centers, except that wage earners tend to live
some distance from the market itself - probably to avoid congestion and pay lower rent. Self-
employed non-farm workers, in contrast, reside near market centers where business is likely to
be better. These findings are consistent with road construction evaluation reports that doc-
ument the rapid increase in non-farm activities and the relocation of villagers along a newly
created road and rural markets (e.g. Rapp 1994, NECMAC 1998, Bajracharya, Aryal, Sharma,
Manandhar and Pyakuryal 1990).
These results are in line with the conclusions of the Government of Nepal (1999). They
also indicates that city size matters: the larger the urban population, the stronger the effect
on non-farm wage employment, indicating a relationship between firm size and city/market
size ((Fafchamps and Shilpi 2001)). Also of interest is the finding that non-farm employment
remains significantly higher in peri-urban areas than in more strictly rural areas. The effect of
town proximity extends for up to four or five hours of travel time away from the town itself -
much further away than normal commuting time.11 This suggests that the development of non-
farm production in rural areas is intimately related to their proximity from large urban centers.
In Nepal, proximity to towns and market centers increases non-farm output, not isolation.
Figures 2 and 3 also present results for self-employment and wage employment in agriculture.
The dependent variable is the share of household labor in these two categories. As anticipated,
the time households devote to farming increases sharply as one moves away from markets and
urban centers. But the pattern varies between self-employment and wage labor: self-employment
"This result is not an artifact of the instrumentation of travel time: virtually identical results are obtained
with actual travel time corrected for road transport.
19



in farming rises as soon as one gets out of the town and remains high. In contrast, wage
employment in agriculture peaks in households residing 1 hour from local markets and in wards
located 5 to 8 hours away from urban centers. This is consistent with the idea that, close to cities,
household with little or no land can find profitable employment opportunities in the non-farm
sector. In more rural wards, the dominant income generation activity for asset poor households
is agricultural wage work. This interpretation is consistent with Jacoby's (2000) finding that
wages fall slightly with distance from the nearest market.'2 An immediate implication is that
land inequality would have a. stronger effect on income inequality and poverty in rural areas
situated far from markets and cities. These areas are also those for which a proper understanding
of agrarian institutions is essential to welfare analysis. When farm and non-farm wage work are
combined, we see a sharp decrease in the share of wage work as one move away from urban
centers. Capitalist modes of production are predominantly an urban phenomenon.
Results regarding migration are less contrasted (Figure 4). The dependent variable is the
share of household members who work outside their ward of residence. We find that work-related
migrations are most prevalent in areas situated 1 to 5 hours away from urban centers, thereby
suggesting that at least part of the non-farm employment encountered in peri-urban areas takes
the form of commuting. The offect, however, is not highly significant. Within a ward, migrations
are concentrated among those households further away from the nearest market, but the effect is
not significant. Taken together, these results suggest that temporary work migration is slightly
more prevalent among households located neither too far nor too close from city influences.
These conclusions are similar to those reached by Seddon et al. (1999) in their work on Nepalese
migration movements.
Turning to farming in more detail, we see that operated land mirrors agricultural employment
12Jacoby, however, does not control for nearby urban population.
20



- it increases with distance from markets and towns (Figures 5 and 6). Farmers within four
hours of a large city have smaller farms than those located in more remote wards. The picture is
slightly different for livestock production. The sale of livestock and livestock products initially
rises with distance from market centers, albeit at a slightly slower pace than cultivated land. But
it eventually tapers off for households located very far from a market center. This suggests that
livestock production as a market-oriented activity is preferred mainly by households residing
between 1 to 4 hours from the local center and more than 4 hours from towns.
Agricultural input use is also influenced by distance. Figures 5 and 6 report a sharp decrease
in fertilizer use as one moves away from cities and market centers - presumably because of
transportation costs.13 Similar results are obtained by Jacoby (2000). But our estimation
method suggests that the effect is not linear. Fertilizer use is highest among households residing
within 2 hours of a market center, after which usage drops sharply.14 This is also the zone within
which crops are grown for the market. These condusions are in close agreement with Jacoby's
(2000) findings that the value of land falls with distance from the market.
Next we investigate the von Thtlnen hypothesis, namely, that what farmers grow varies
systematically with distance to towns and markets. For that purpose, the share of land planted to
various categories of crops is used as dependent variable. Two cropping seasons are distinguished
- the wet season when paddy is produced, and the dry season when the focus is on other cereals.
The regressions are conditional on the household producing crops, an activity that, as we have
seen, takes place outside towns and cities. Some of the results are presented graphically in
Figures 7 and 8 for f(.) and g(.), respectively. We find that vegetable production, as predicted,
is highest in the vicinity of cities and local markets, both during the dry and wet season. The
13A similar regression conducted on fertilizer price indeed shows a sharp price increase with distance from cities.
Households located within the same ward all pay the same price for fertilizer, but have to haul it over different
distances.
'4Virtualy identical results obtain if one uses fertilizer per cultivated area.
21



magnitude of the effect is large but it is seldom significant. Vegetable production also increases in
wards located far from towns, something we had not anticipated. Contrary to results repcorted so
far, spatial cropping patterns are quite different when actual population and travel time figures
are used. In that case, we fincl a stronger and clearer relationship between vegetable production
and proximity to towns. The difference suggest a tendency for Nepalese roads to be built to
serve vegetable producing areas.
Land devoted to paddy initially increases with distance from the local market. This effects,
however, tapers off in more reraote parts of the ward. There, more emphasis is put on subsistence
crops. Areas located near wrban centers produce slightly more paddy, but the effect is not
significant. Here too, the difl9erence with uninstrumented regressions is large, suggesting that
roads are built to paddy producing areas. During the dry season, pulses, oilseeds, and other
(minor) crops follow a pattern similar to that of paddy, albeit less pronounced. These results are
in line with the concentric circles hypothesis. Perishable crops such as vegetables are produced
near markets and cities. Less perishable commercial crops appear in areas further away from
the city, but not so far that transport becomes problematic. In areas more than 8 hours away
from cities, we observed a decreased emphasis on commercial crops and a switch from paddy to
other cereals. These changes reflect the self-subsistence focus of isolated areas.
To confirm our results, we examine whether the commercialization of crops varies systemat-
ically with distance. To this elfect, we regress the proportion of total agricultural output that is
sold to the market (Figure 9).1) By construction, the regression is limited to farming households.
Results show a lot of variation in crop commercialization initially among households located near
markets and towns. As Jacoby (2000), we find that commercialization initially increase with
the distance between households and local markets, but the effect is not significant. Up to 6
1Regressions for individual crops yield similar results.
22



hours of travel time, distance from towns has the expected monotonic effect on crop sales: more
distant wards sell less. Crop sales are zero on average for households residing 6 hours or more
from a large town, an indication of the self-subsistence focus of agricultural production.
Next we turn to consumption. We first regress consumption shares in major food products
on distance to markets and cities. Taken together, these products account for 86% of measured
consumption. Results are sununarized in Figure 10. Rice consumption tends to fall with distance
from markets, but proximity to towns has no significant effect. Among households residing more
than two hours away from the nearest market spend roughly 25 percentage points less on rice, the
difference going to cereal consumption. This pattern is probably related to the need to process
paddy into rice before consumption. Indeed, rice milling facilities are typically located in and
around markets. In contrast, consumption of other crops increases with distance from markets.
Livestock expenditures, in contrast, show no clear pattern. Proximity to urban centers is shown
to raise the consumption of livestock products (nearly significant) and reduce the consumption
of cereals other than rice. Contrary to expectations, consumption patterns in isolated wards
resembles more that of peri-urban areas than that of intermediate wards. We revisit this issue
in the next section.
The next step is to investigate whether differences in consumption are related to crop mon-
etization. We regress the share of cash purchases in total consumption for fouu agricultural
products for which we have data: rice; other cereals; vegetables; and fruits. Results for all crops
combined are summarized in Figure 11. They indicate a systematic decline of the share of pur-
chased consumption with distance: households residing near a town or market are more likely to
purchase the food they consume. This hardly comes as a surprise, given that urban households
are less likely to farm and thus to produce their own food.16 The effect is large in magnitude
"The question is whether this is the only reason. We revisit this issue in the next section.
23



both within wards and between wards and towns; it is very significant in all cases. Resi,lents of
wards located less than 3 hours away from a town rely massively on the market to fulfill their
consumption needs. In contrast, households living more than four hours from a city do not at all
rely on the market for their consumption of agricultural products. Taken together with earlier
findings about the sale of crol) output, these results confirm that crop commercialization - both
in terms of sales and purchasE s - is linked to proximity to markets and urban centers.
Before we close this section, we examine how sale and consumption pattern combine to
generate a spatial distribution of marketed surplus. We use two definitions of marketed surplus:
first, as the ratio of crop production over consumption;17 second, as the difference between
production and consumption. The first definition emphasizes self-sufficiency, that is, the degree
to which a household is in surplus or deficit, irrespective of the size of its output. The second
definition emphasizes the absolute size of the marketed surplus. Results, presented in Figure 12,
show that food self-sufficiencyr increases rapidly with distance from towns and markets. Given
that all households are included in the regression, this outcome largely reflects the fact that
households located near cities and market centers spend less time producing food and more time
producing non-farm products and services.
In terms of absolute magnitude, marketed surplus is largest at intermediate distances. Mar-
keted surplus is largest one hour or more from a market center and four hours from towns; the
latter effect, however, is not significant. Given the nature of the terrain in Nepal, transporting
crop surpluses to the market is a non-negligible task. It is therefore not surprising that the value
of the marketed surplus drops among households residing more than two hours from a market.
Noticeable for all crops, this outcome stands in contrast with our previous result for livestock,
which showed that the sale of livestock products only begins to drop among households located
17Because the resulting ratio has a highly skewed distribution, we use the log(l+ ratio) instead as dependent
variable.
24



more than four hours from a local market. Taken together, these results suggest that in the
2-4 hour range, livestock progressively supplants crops as the main cash generating activity for
farmers.
To summarize, we have seen that non-farm activities are concentrated in and around cities
and markets. We were able to show that the range of distance over which non-farm wage
employment is stimulated by cities is much larger than what is often believed. Contrary to
claims occasionally made, we find no evidence that cities eliminate non-farm employment in
their hinterland - albeit they may eliminate certain activities.18 In terms of agriculture, several
forces are at work. The dominant force is that which affects commercialization. Proximity to
towns and markets has a strong positive effect on the sale and purchase of agricultural products.
The commercialization of agriculture is thus predominantly a peri-urban phenomenon. This
finding is consistent with the fact that non-farm production is also higher in peri-urban areas:
villagers who do not farm rely on the market for their consumption, and they can afford to
rely on the market because traded quantities are higher. Fertilizer purchases follow the same
pattern, suggesting that technology adoption is related to crop commercialization.
The effect of proximity is felt over a four hour radius around cities. Beyond this distance,
crops are rarely sold and purchased. Livestock products, however, continue to be marketed
well beyond this limit, suggesting a geographical pattern of specialization with crops close to
cities and livestock further away. This is consistent with the fact that livestock products either
transport themselves (live animals) or have a high value to weight ratio (ghee). Within the zone
of influence of cities, agricultural specialization also varies with distance, with vegetables being
produced closest to markets and paddy, pulses, and oilseeds produced further away. The effect,
however, is not as strong as anticipated. Concentric circles of specialization in Nepal nevertheless
l8This issue is the object of future work.
25



seem to fit the von Thunen hypothesis. Our results also demonstrate the usefulness of the non-
parametric approach: with few exceptions, the effect of distance is highly non-linear, with urban
areas and isolated villages sometimes presenting similar features. The non-parametric approach
also proved useful in identify.rng the precise spatial range of particular phenomena.
5. Controlling for Household Characteristics
We have seen that what Nepalese households do varies systematically with proximity to towns
and market, but we do not know why. Our preferred interpretation is that what people do is
influenced by market access, but other explanations are possible as well. In particular, prefer-
ences or talent may vary systematically with proximity to markets in such a way as to generate
the relationships described in the previous section. Wage work, for instance, might be more
frequent in the vicinity of tovns simply because people there are better educated and edlucated
workers are more productive in a salaried workplace.
To show this formally, consider a producer in location j having to choose among various
activities indexed by i. Let the output price be denoted pij.19 There is a single production
factor, labor. Output Qij of activity i depends on employment Lj in activity i, with Q' < 0.
Profit maximization dictates that marginal value product of labor equal the wage rate wj in
location j, i.e., that:
PijQ'(Lij) = wj
from which we see that the optimal choice of Lij increases with pij: if output prices pij for certain
activities are higher closer to cities and markets, so is Lij. More labor is used in remunerative
activities.
l9What matter for activity choice are only relative prices. Prices are normalized by setting one price to one.
26



Now consider an alternative model in which output depends on two factors of production,
labor Lij and human capital Hij. Further assume that relative output prices are the same
across locations. Factor prices are written wj for labor and rj for human capital. Relative factor
prices depend on their relative scarcities: locations with a lower HjlLj have a higher rj/wj.
Comparative advantage dictates that locations specialize in the activities using more intensively
the factor they have in abundance. So, for example, if activity k uses more human capital and
human capital is abundant in location j, labor used in activity k will be higher in locations with
a higher human capital, i.e., Lij is an increasing function of Hj. Further assume that Hj varies
systematically with proximity to town and market, say, because schools are located near towns
and market centers. Regressing Lij on proximity to town and market, as we have done so far,
would show a relationship between Lij and distance even in the absence of differences in relative
prices, i.e., in the absence of market effects.
The above discussion suggests an indirect way of testing whether patterns of production
vary because of differences in factor endowments or because of differences in marginal returns
possibly driven by proximity to markets. Say we have data on factor endowment Hj. First
regress Hj to ascertain whether it varies systematically with distance. If it does, differences
in factor endowment might be responsible for the measured effect of distance on activity. To
test this possibility, regress Lij on Hj as well as distance: if including Hj drives the effect of
distance to zero, factor endowment differences might be responsible for the observed spatial
division of labor. If distance remains significant, distance affects marginal returns through
some other process, possibly relative prices, possibly other unobserved endowment. A similar
reasoning can be extended to choices other than production. Consumption, for instance, may
vary systematically across locations not because relative prices are different but because incomes
vary with location.
27



Implementing the above approach is the object of this section. Before presenting the results,
however, it is important to understand its limitations. First of all, we can only test the effect
of observed endowment diffarences. In the absence of panel data, unobserved differences such
as activity specific skills or preferences cannot be controlled for.20 Second, the validity of the
test rests on Hj being exogenous. Suppose that this is not the case, i.e., that factors are either
perfectly mobile across space or that they can be accumulated instantaneously, so that factor
prices are equalized across space. Further assume that prices vary systematically with location.
Profit maximization dictates that:
Pij 9Q(Lj,")    =   r
Pi  Hi,
which implies that Lij, Hij, and Hj =   Hij vary systematically with Pij and thus with distance.
The endogeneity of Hj generates multicollinearity with distance since Hj depends on relative
prices and by assumption Pij = EJ T1kDk + EhH= 'Yhph +gij where eij represents factors other
than distance that affect relative prices - and thus Lij and Hj. When Eij is large, Hj is much
more correlated with Lij than distance itself. Including Hj in the regression thus biases the
distance coefficients toward zero.
Since we do not have instruments for endogenous placement of households, we cannot con-
trol for the possible endogerLeity of Hj. We nevertheless propose to do two things: (1) regress
household characteristics Hj on distance and see whether Hj vary systematically with distance;
and (2) add Hj to all regressions presented in section 4. Given that Hj is potentially endoge-
nous, it is important to understand the limits of this procedure. If distance coefficients become
20Plans are underway to collect a second round of NLFS panel data in Nepal but the data will not be available
for several years.
28



non-significant when we include household characteristics, it could be because the location of
endowments drives economic activity, or because households locate endogenously. We cannot
tell. If, however, distance is still significant after we include household characteristics, then
we can reject the hypothesis that differences in observed factor endowments are responsible for
spatial division of labor. If, in addition, observed factor endowments or preference shifters do
not vary with distance, then they can safely be ruled out as the driving force behind spatial
division of labor.
The household characteristics used in the analysis are measures of human and physical
capital for which we have data - age of household head, household size, household composition
by sex and age, average education of adult males and females, and inherited wealth. The age
of household head controls for experience and life cycle effects. Household size and composition
control for labor supply and food preference effects.2' Average years of schooling for adult males
and females our measure of human capital. Children education is ignored because it is less likely
to affect activity choices but more likely to be influenced by it through income. Inherited wealth
is our proxy for wealth. Measures of current wealth such as land and livestock cannot be used
because they depend on current activity choice (farming). To control for other geographical
effects, we also include regional dummies (east, west, mid-west, far-west, and central - the
omitted category) and elevation dummies (mountains, hills, and Terai - the omitted category).
To control for income effects, regressions relative to consumption (expenditures on certain food
categories; food purchases; and marketed surplus) also include total expenditures as additional
regressor.
We begin by examining whether household characteristics vary systematically with distance
21 Household composition variables are the share of adult males, the share of adult females, the share of children,
and the share of youths (the omitted category).
29



to market and towns. Results are summarized in Figures 13 and 14.22 Households residing close
to markets have fewer children and more adult males. Household heads residing close to towns
tend to be younger. As expected, schooling levels of male and female adults drop with distance
from towns and markets. This could be due either to supply effects - school are located closer
to market center - or demand effect. The latter would arise if higher returns to schooling near
the market center incites eclucated residents and migrants to locate closer to the market and
encourages parents to invest in their children's education.
We also examine whether inherited wealth and total consumption vary systematically with
distance. Results, presented in Figure 15, show that inherited wealth increases dramatically
with distance from towns and markets. This is consistent with the emphasis on agriculture and
the role of land and livestock in agricultural production. The lower half of Figure 15 also shows
that consumption drops dramatically with distance, albeit it is unclear how much of this drop
simply reflects lower food prices in rural areas.
Observed household characteristics vary with distance. Are results reported in the previous
section ultimately driven bv differences in endowments and preferences? To investigate this
possibility, we estimate regressions of the form:
J         H
y =H +H    rjD+E7hPi +                                (5.1)
j=1       h=1
where Hi stands for household characteristics and regional dummies. The model is estimated
using the roughness penalty correction as before. To save space, we omit the Figures themselves
and simply discuss the diffeiences with Figures 2 to 12.
Our first set of results relate to non-farm work. They can be summarized as follows. The
22Some household characteristics are omitted from the figures because non-significant but included in the
subsequent analysis.
30



effect of distance from urban population is reduced by about 30% by adding Hi but distance
coefficients remain significant if they already were. We can therefore rule out the possibility that
the effect of proximity towns is due to differences in observed household characteristics. Within
wards, however, distance is much less significant. One possibility is that rural markets locate
where educated households reside. A more likely interpretation is that residential mobility and
reallocation of productive resources within wards are sufficient to match workers with location-
specific occupations. This would be the case, for instance, if households interested in trade or
crafts move to the market center. This would be in line with field observations made by Rapp
(1994) and Bajracharya et al. (1990). Without more data, however, we cannot be more conclusive
on this issue. The effect of proximity to towns on the propensity to migrate is unchanged.
Next we turn to farming. Results concerning operated land, livestock, fertilizer use, cropping
patterns, and crop sales (Figures 5 to 9) are by and large unaffected as far as distance to towns is
concerned. But distance to the local market is no longer significant in nearly all cases. The key
determinants are education and inherited wealth (which determines the start-up capital available
for farming): these regressors account for most of the within-ward variation in farming as more
educated households farm less, while those with more inherited wealth farm more. These results
are consistent with Fafchamps and Quisumbing (1999) and Yang (1997).
We also reestimate our consumption, crop purchases, and marketed surplus regressions (Fig-
ures 10 to 12). Since consumption choices depend on income, consumption shares and crop
purchases are likely to be affected by income effects which, in turn, are affected by activity
choices. As a result, market participation results presented in the previous section may be
spurious because driven by income differentials.
To control for income effects, we add total consumption expenditures (in log) and its square
to the list of household characteristics. Results indicate that distance to local market is no
31



longer significant in all three sets of regressions. Household characteristics, especially consump-
tion expenditures, account for observed within-ward differences. The situation is quite differ-
ent for distance to towns: results here are stronger than before the introduction of household
characteristics. As before, crcp purchases decrease with distance from towns and, as a result,
self-sufficiency increases. The effect remains strong and significant. The consumption of farm
products depends on distance to towns, but in ways that become clearer after income effects
are controlled for. This is shcwn in Figure 16. In particular, rice consumption is much higher
near towns, while the consumption of other crop and livestock products increases with distance
to towns. This is in line WithL studies that have brought to light the role of rice as an urban
food, possibly because of ease of preparation (e.g. Savadogo and Brandt 1988, Nagy, Sanders
and Ohm 1988).
6. Conclusion
Combining household level data with information on distance to markets and towns, we have
examined the spatial division of labor in Nepal. A semi-parametric model was used to avoid
restricting the shape of the relationship between spatial specialization and distance to markets
and towns. Distance is measured in hours of travel time. To control for possible endogeneity,
city population and travel time are instrumented using physical characteristics of surveyed wards
and districts. Coefficient estimates with and without instrumentation are very similar, thereby
suggesting that endogeneity of city size and road construction is not a serious concern.
Result indicate a strong spatial division of labor. Non-farm employment - our indicator of
non-farm production - is heavily concentrated in markets and in and around towns. The effect
is strong for towns of all sizes. Agricultural wage employment, in contrast, is concentrated in
rural areas sufficiently close to cities that they can specialize in commercial crops but neither so
32



close that non-farm employment takes over, nor so far that they revert to self-subsistence.
Crop choices vary with distance to town and with their size. Vegetable production in both
seasons is somewhat concentrated at the vicinity of markets and urban centers while, paddy,
pulses, oilseeds, and other commercial crops are more important at intermediate distances.
Market participation varies with distance as well. Households near markets and cities buy most
of the rice and agricultural products they consume, even when we control for land and other farm
assets. These households also sell a larger proportion of their crop production. An examination
of the spatial distribution of marketed surplus indicates that food for urban consumption is
mainly produced in an intermediate zone located 30 minutes to 2 hours away from the nearest
market and 3 to 7 hours from toWIIs. Beyond this zone, households derive some cash income
from the sale of livestock products, but this phenomenon itself tapers off beyond 8 hours of
travel time from the nearest market.
Surnmarizing our results, the proximity to markets and size of towns are strongly associated
with different patterns of production among Nepalese households. Towns themselves specialize
in the production of non-farm products. They buy vegetables from immediately surrounding
areas and cereals from villages located a little further away. Agricultural marketed surplus is
produced in an intermediate zone where much agricultural wage work is also found. Surprisingly,
this is not the zone where fertilizer consumption is highest.
Isolated households and villages essentiaLly rely on self-subsistence, both in terms of non-farm
production and food self-sufficiency. Broadly consistent with earlier findings by Jacoby (2000),
all these findings fit the von Thurnen model of concentric specialization, except that our results
also show the importance of town size. Less anticipated is our finding that villages located
near towns also participate in non-farm production and that the proportion of agricultural wage
work only rises slowly with distance from cities. These suggest that proximity to cities is not
33



detrimental to non-farm production, contrary to what is sometimes assumed. This appears to
be linked not so much to income effects but rather to reliance on the market for the satisfaction
of consumption needs.
These results are partly associated with differences in factor endowments and preferences.
Distance between surveyed households and the nearest market is no longer significant in most
regressions once we control for household characteristics. Without household panel data, we
are unable to tell whether i!t is proximity to market that drives household factor endowments
(through internal migration or accumulation) or whether it is the opposite. Distance to towns,
however, remains strongly significant even after household characteristics are included. This sug-
gests that distance to towns has an effect on economic activity and market participation beyond
that of differences in (observed) factor endowments. These issues deserve further research.
References
Abdel-Rahman, Heshani M., "Product Differentiation, Monopolistic Competition and City
Size," 1988, 18, 69-86.
, "Productivity Gains from Geographic Concentration of Human Capital: Evidence from
the Cities," Journal oJ Urban Economics, 1993, 34, 380-400.
,"Economies of Scope in Intermediate Goods and a System of Cities," Regional Science
and Urban Econ., 1994, 24, 497-524.
Arthur, W. Brian, "Sell-Reinforcing Mechanisms in Economics," in "The economy as an
evolving complex system, SFI studies in the sciences of complexity," P.W. Anderson, K.J
Arrow and D. Pines (eds.), 1988.
34



Bajracharya, Bhuvan B., Sushil B. Aryal, Pitamber Sharma, Mangal S. Manand-
har, and Kailash N. Pyakuryal, Impact of Lamosangu-Jiri Road and Integrated Hill
Development Programmes on the lhansport and Food Sectors in Dolakha and Sindhupal-
chowk Districts, Kathmandu, Nepal: Centre for Economic Development and Administra-
tion, Tribhuvan University, 1990. Prepared for the Swiss Development Cooperation.
Barrett, Christopher B. and Paul A. Dorosh, "Farmers' Welfare and Changing Food
Prices: Nonparametric Evidence from Rice in Madagascar," American Journal of Agricul-
tural Economics, 1996, 78, 656-669.
Barro, Robert J. and Xavier Sala-I-Martin, "Convergence across States and Regions,"
Brookings Papers on Econ. Activity, 1991, 1, 107-158.
Baulch, Bob, "Transfer Costs, Spatial Arbitrage, and Testing for Food Market Integration,"
American Journal of Agricultural Economics, 1997, 79(2), 477-487.
Chomitz, Kenneth M. and David A. Gray, "Roads, Land Use, and Deforestation: A
Spatial Model Applied to Belize," World Bank Economic Review, 1996, 10(3), 487-512.
Christaller, W., Central Places in Southern Germany, Englewood Cliffs, NJ: trans. C.W.
Baskin, Prentice-Hall, 1966. (Originally published 1933).
Ciccone, Antonio, "Dynamic Externalities and the Spatial Distribution of Wages in the US,"
1997. (mimeograph).
and Robert E. Hall, "Productivity and the Density of Economic Activity," Amer. Econ.
Rev., 1996, 86(1), 54-70.
Conley, T.G., "GMM Estimation with Cross-Sectional Dependence," Journal of Econometrics,
1999, 92(1), 1-45.
35



Dercon, Stefan, "On Market Integration and Liberalisation: Method and Application to
Ethiopia," Journal of Development Studies, 1995, 32(1), 112-143.
Desmet, Klaus, "A Perfect Poresight Model of Regional Development and Skill Specialization,"
1998. (mimeograph).
and Marcel Fafchamnps, "Agglomeration and Congestion in U.S. Counties," 2000.
(mimeograph).
Dicken, Peter and Peter E. LLoyd, Location in Space: Theoretical Perspectives in Economic
Geography, New York: H:arper Collins, 1990.
Dore, Ronald, Taking Japan Seriously: A Confucian Perspective on Leading Economic Issues,
Stanford: Stanford University Press, 1987.
Ellison, Glenn and Edward L. Glaeser, "Geographic Concentration in U.S. Manufacturing
Industries: A Dartboard Approach," Journal of Political Economy, 1997, 105(5), 889-927.
Fafchamps, Marcel, "The Enforcement of Commercial Contracts in Ghana," World L)evelop-
ment, 1996, 24(3), 427-448.
- and Agnes Quisumbing, "Human Capital, Productivity, and Labor Allocation in Rural
Pakistan," Journal of Human Resources, 1999, 34 (2), 369-406.
-     and Brigit Helms, "Local Demand, Investment Multipliers, and Industrialization: The-
ory and Application to the Guatemalan Highlands," J. Devel. Econ., 1996, 49, 61-92.
-     and Forhad Shilpi, "Cities and Specialization: Evidence from South Asia," 2001.
(mimeograph).
Fujita, Masahisa, "A Monopolistic Competition Model of Spatial Agglomeration," Regional
Science and Urban Econ., 1988, 18, 87-124.
36



, Paul Krugman, and Anthony J. Venables, The Spatial Economy: Cities, Regions,
and International Trade, Cambridge and London: MIT Press, 1999.
_- , and Tomoya Mori, "On the Evolution of Hierarchical Urban Systems," European
Economic Review, 1999, 43(2), 209-251.
Glaeser, Edward L., Hedi D. Kallal, Jose A. Sheinkman, and Andrei Shleifer,
"Growth in Cities," J. Polit. Econ., 1992, 100(6), 1126-1152.
_, Jose A. Scheinkman, and Andrei Shleifer, "Economic Growth in a Cross-Section of
Cities," Journal of Monetary Economics, 1995, 36(1), 117-143.
Good, I.J. and R.A. Gaskins, "Nonparametric Roughness Penalties for Probability Densi-
ties," Biometrika, 1971, 58, 255-277.
Government of Nepal, Nepal Agricultural Perspective Plan (Final Report) - Main Document,
Kathmandu, Nepal: National Planning Commission, His Majesty's Government of Nepal
and Asian Development Bank, 1995. Prepared by the Agricultural Projects Services Centre,
Kathmandu and John Mellor Associates, Washington D.C.
, Report on the Nepal Labour Force Survey 1998/99, Kathmandu, Nepal: Central Bureau of
Statistics, National Planning Commission Secretariat, His Majesty's Government of Nepal,
1999.
Hall, Robert E. and Charles I. Jones, "The Productivity of Nations," 1996. (mimeograph).
Harris, John and M. Todaro, "Migration, Unemployment and Development: A Two-Sector
Analysis," Amer. Econ. Rev., 1970, 60, 126-142.
Henderson, J. Vernon, Urban Development: Theory, Fact, and ilusion, New York: Oxford
University Press, 1988.
37



Henderson, J.V., "The Sizes and Types of Cities," Amer. Econ. Rev., 1974, 64(4), 640-656.
Hununels, David, "Essays on Spatial International Economics," 1995. (unpublished Ph.D.
thesis).
Isard, W., Location and Slace Economy, Cambridge, Mass.: MIT Press, 1956.
Jacobs, Jane, The Economy of Cities, New York: Random House, 1969.
-, Cities and the Wealth of Nations, New York: Random House, 1984.
Jacoby, Hanan G., "Access to Markets and the Benefits of Rural Roads," Economic Journal,
2000. (forthcoming).
Krugman, Paul R., Geography and 7rade, Leuven and Cambridge, Mass.: Leuven U.P. and
MIT Press, 1991.
"Increasing Returns and Economic Geography," J. Polit. Econ., 1991, 99(3), 483-99.
Lampard, Eric E., "The Evolving System of Cities in the United States: Urbanization and
Economic Developmenl;," in "Issues in Urban Economics," Baltimore: Harvey S. Perloff
and London Wingo (eds.), John Hopkins Press, 1968.
Lee, Tae Jeong, "A Spatial Pattern of Household Specialization in the Marketing Sys-
tem of Peasant Econoraies: With Evidence from Oaxaca Valley of Mexico, and Western
Guatemala," 1993. (mimeograph).
Lewis, W. Arthur, "Economic Development with Unlimited Supplies of Labour," The Manch-
ester School, 1954, XXH(2), 139-191.
Losch, A., The Economics of Location, New Haven, Conn.: Yale University Press, 1954. (Orig-
inally published 1939).
38



Minten, Bart and Steven Kyle, "The Effect of Distance and Road Quality on Food Collec-
tion, Marketing Margins, and Traders' Wages: Evidence from the Former Zaire," Journal
of Development Economics, 1999, 60, 467-495.
Nagy, J.G., J.H. Sanders, and H.W. Ohm, "Cereal Technology Interventions for the West
African SAT," Agricultural Economics, 1988.
NECMAC, Traffic Survey Along Benighat-Aarughat/Arkhet Road (Final Report), Kathmandu:
Impact Monitoring Unit, GTZ, 1998.
Radelet, Steven and Jeffrey Sachs, "Shipping Costs, Manufactured Exports, and Economic
Growth," 1998. (mimeograph).
Rapp, Elke, Impact Assessment of the Local Road Program Implemented on Dhading Besi-
Salyantar-Siktar Road and Bhimbhunga-Lamidanda Road, Kathrnandu: Impact Monitoring
Unit, German Technical Cooperation, 1994.
Rauch, James E., "Productivity Gains from Geographic Concentration of Human Capital:
Evidence from the Cities," Journal of Urban Economics, 1993, 34, 380-400.
Ravallion, M., "Testing Market Integration," Amer. J. Agric. Econ., 1986, 68(1), 102-109.
Reardon, Thomas, "Using Evidence of Household Income Diversification to Inform Study of
the Rural Nonfarm Labor Market in Africa," World Development, 1997, 25(5), 735-747.
Savadogo, Kimseyinga and Jon A. Brandt, "Household Food Demand in Burkina Faso:
Implications for Food Policy," Agricultural Economics, 1988, pp. pp. 345-.
Seddon, David, Jagannath Adhikari, and Ganesh Gurung, "Foreign Labour Migration
and the Remittance Economy of Nepal," 1999. (mimeographed report to DfID).
39



Silverman, B.W., "On thE Estimation of a Probability Density Function by the Maximum
Penalized Likelihood Method," Annals of Statistics, 1982, 10, 795-810.
, "Spline Smoothing: The Equivalent Variable Kernel Method," Annals of Statistics, 1984,
12, 898-916.
Takayama, Takashi and George G. Judge, Spatial and Temporal Price and Allocation
Models, Amsterdam: North-Holland, 1971.
Timmer, C. Peter, Getting Prices Right: The Scope and Limits of Agricultural Price Policy,
Ithaca: Cornell University Press, 1986.
Venables, Anthony J. and Nuno Limao, "Geographical Disadvantage: a von Thunen Model
of International Specialization," 1999. (mimeograph).
von Thunen, Johann Heinrich, Isolated State: an English Edition of Der Isolie0te Staat,
Oxford: Pergamon Pre3s, 1966.
Yang, Dennis T., "Education and Off-Farm Work," Economic Development and Cultural
Change, 1997, 45 (3), 613-632.
40



Table 1. Characteristics of Surveyed Households
1. Household composition             Unit   Mean Std. dev.  Median   N. obs.
Household size, of which:     Number       5.6      2.8       5     3344
Adult males                 Number        1.2      0.8       1    3344
Adult females               Number        1.3      0.8       1    3344
Teenagers                   Number        1.9      1.6       2    3344
Children                    Number        0.9      1.0       1    3344
Elderly                     Number        0.2      0.5       0    3344
2. Human capital
Age of household head         Years         44       15      43     3344
Age of spouse                 Years         39       12      37     3344
Years of schooling of adult males  Years    0.9     1.5       0     3344
Years of schooling of adult females Years   0.4     0.9       0     3344
3. Land
Total operated area, of which:  Ha          0.8     1.7      0.4    2653
Irrigated land              Ha            0.3      0.8       0    2653
Total owned area              Ha            0.8     1.7      0.3    2653
4. Assets
Cows and buffaloes            Number        2.6     3.1       2     3345
Farm equipment                Rupees      1549    20948     120     2220
5. Location
Distance from hh to nearest market Hours    2.2     3.4      1.1    3344
Distance from ward to nearest town Hours    4.0     4.7      2.2    3344
41



Table 2. Production and Consumption
1. Labor                                      Unit          Mean     Std. dev.   Median  N. obs.
Share of total household employmenl. In:
Non-farm employment                        Share        30.2%       37.6%     10.5%     3248
Non-farm self-employment                 Share         12.1%      27.5%       0.0%     3248
Non-farm wage employment                 Share         18.1%      30.3%       0.0%     3248
Farm employment                           Share         69.8%       37.6%     89.5%     3248
Farm self-employment                     Share         56.7%      38.1%      65.0%     3248
Farm wage employment                     Share         13.1%      24.3%       0.0%     3248
Wage employment                            Share         31.3%      34.6%     19.2%     3248
Self-employment                            Share         68.7%      34.6%     80.8%     3248
Share of household members who work outside village Share    4.2%      11.1%      0.0%     3337
2. Agriculture and livestock
Operated area                              Ha             0.81        1.67      0.37    3337
Value of annual livestock sales            Rupees         2937       7529          0    3337
Value of annual sale of livestock products  Rupees        1202       4207          0    3337
3. Cropping pattern (farmers only)
Share of planted acreage In:
Wet season:
Paddy                                    Share         38.2%       35.0%     31.0%     2377
Other Cereals                            Share         33.7%       30.9%     29.6%     2377
Pulses                                   Share         11.2%       15.7%      0.0%     2377
Vegetables                               Share          9.6%       16.4%      0.0%     2377
Other crops                              Share          7.3%       15.6%      0.0%     2377
Dry season:
Cereals                                  Share         53.2%       34.4%     55.3%     2204
Pulses                                   Share         11.3%       19.1%      0.0%     2204
Oilseeds                                 Share         11.5%       18.7%      0.0%     2204
Vegetables                               Share         16.8%       25.0%      4.2%     2204
Other crops                              Share          7.2%       15.8%      0.0%     2204
4. Agricultural Input use (farmers only)
FerUlizer quantity                         Kg               70        144         20    2329
Fertilizer per area                        Kg/ha           125        440         33    2322
5. Sale of agricultural products (producers only)
Sales of crop as % of value of crap output:
Paddy                                    Share          9.3%       18.4%      0.0%     1989
Other cereals                            Share          9.4%       19.5%      0.0%     2358
Other crops                              Share         14.4%       25.0%      0.0%     2252
All crops combined                       Share         12.5%       18.9%      0.0%     2505
S. Consumption of agricultural products
Share in total consumption:
Rice                                       Share         29.8%      10.5%      30.9%    3337
Other cereals                              Share         47.3%      11.9%      47.5%     3337
Vegetables                                 Share          7.2%       5.6%       5.8%     3337
Fruits                                     Share          1.9%       3.5%       1.1%    3337
Share of cash purchases in consumptlon of:
Rice                                       Share         52.4%      41.8%     51.7%     3326
Othercereals                               Share         36.1%      41.1%      15.9%     3206
Vegetables                                 Share         43.1%      40.2%      31.4%     3317
Fruits                                     Share         58.3%      43.5%      75.6%    3058
All combined                               Share         44.2%      36.4%      37.5%     3337
7. Marketed surplus
Value of output minus value of consumption
Paddy/rice                                 Rupees        -1683      13615      -2660    3337
Other cereals                              Rupees          545      27248       -240    3337
Vegetables                                 Rupees        -1356       3032       -950    3337
Fruits                                     Rupees         -257       4626       -192    3337
All agricultural products combincid        Rupees         -650      34704      -3538     3337
42



Table 3. Distance to the Nearest Town
Number of          Number of
7. Nearest City wlthin       Wards      Percent  Households
0 -1 hour of travel time        80    29.2%        951   28.4%
1 - 2 hour of travel time       52    19.0%        634   19.0%
2 - 3 hour of travel time       40    14.6%        485   14.5%
3 - 4 hour of travel time       16     5.8%        199    6.0%
4 - 5 hour of travel time       10     3.6%        117    3.5%
5 - 6 hour of travel time       12     4.4%        152    4.5%
6- 7 hour of travel time         9     3.3%        108    3.2%
7 - 8 hour of travel time        6     2.2%         72    2.2%
8 - 9 hour of travel time        9     3.3%        115    3.4%
9 - 10 hour of travel time       7     2.6%         84    2.5%
10 - 11 hour of travel time      6     2.2%         75    2.2%
11 - 12 hour of travel time      6     2.2%         80    2.4%
12 - 13 hour of travel time      7     2.6%         98    2.9%
13 - 14 hour of travel time      1     0.4%         12    0.4%
14 - 15 hour of travel time      4     1.5%         49     1.5%
15 - 30 hour of travel time      9     3.3%        113    3.4%
274                3344
43



Table 4. Proximity to Urban Population
Mean    Std. dev. Median
City population within:
0 -1 hour of travel time  74804   180351       0
1 - 2 hour of travel time  54514  130480       0
2 - 3 hour of travel time  97443  167273   28778
3 - 4 hour of travel time  151752  223457  69968
4 - 5 hour of travel time  164544  225559  103689
5 - 6 hour of travel time  214631  268668  132280
6- 7 hour of travel time  342046  320305  245629
7 - 8 hour of travel time  338632  320061  296365
8 - 9 hour of travel time  288493  299774  188042
9 - 10 hour of travel time  341247  309812  284257
44



Appendix 1: Intrumenting equation for town population
(dependent variable is the log of town population)
unit              coef.  t-stat
Area of district in which town is located  log(square km)  -0.472  -1.38
Total arable land area                log(ha)           -0.592  -1.18
Distance to nearest navigable part of river  log(km)    -0.206  -1.04
Mean elevation of the district        meters            -0.000  -0.04
Standard deviation of district elevation  meters         0.001   0.97
Mountainous terrain                   yes=1             -1.218  -1.85
Intercept                                               21.034   3.63
Number of observations                                        34
R-squared                                                  0.2705
Source: distance to river, elevation, and area of district come the Nepal GIS data base.
Arable land comes from the Statistical Yearbook of Nepal.
45



Appendix 2: Instrumenting equation for travel time
(dependent variable is the log of travel time between ward and town)
unit              coef.  t-stat
Foot travel time between ward and town  log             0.677  174.12
Ward characteristics
Area of district in which ward is located  log(square km)  0.202  21.65
Total arable land area of district   log(ha)           -0.145  -12.82
Mean elevation                        meters            0.000   14.02
Standard deviation of district elevation  meters       -0.000   -6.71
Central region                        yes=1            -0.229  -24.06
West region                           yes=1            -0.281  -26.65
Mid-west region                       yes=1            -0.219  -14.90
Far-west region                       yes=1            -0.199  -11.98
Town characteristics
Area of district in which ward is located  log(square km)  0.067  5.34
Distance to nearest navigable part of river  log(km)    0.048    7.28
Total arable land area of district    log(ha)          -0.005   -0.30
Mean elevation                        meters            0.000    8.18
Standard deviation of district elevation  meters       -0.000   -6.35
Central region                        yes=1            -0.150  -11.40
West region                           yes=1             -0.186  -15.86
Mid-west region                       yes=1            -0.172   -9.55
Far-west region                       yes=1            -0.082   -4.20
Intercept                                              -1.780   -9.37
Number of observations                                      9187
R-squared                                                 0.8443
Source: foot travel time computed from GIS data correcting for elevation.
Other variables as in Appendix 1. Eastem region is the omitted region.
46



Figure 1. Map of Nepal and Location of Surveyed Villages
CI CJ
,4 'V' /'                            Urban centers
,- 1-29C5O
A ,t ;  <  -~-  ),  *                 U)3 123$9L8'5
.   ,   t_*       >            >  1533- .--,  4-,
tLf S12  A   :    i V <~~~~12W9t2"2%
p. P14 +     U. +-*' -4-.     *



0.38 -0.47-
0   2O0
0,0
a,
-0.003-                                         0.00j_
.t2   35stance from local market in hours (log scaleT   .stance frbm local market in hours (log scale)
Non-farm wage employment                        Non-farm self-employment
0Fg0.23 -                                      0.84 -
00
0                                  ~~~~~~~~~~~~~~~0
a,~~~~~~~~~~~~~~~~~~~~~~~~~~~~~I
U,                                             U
-0.03                                          0.00
.12                     2           ~          12sae
cpistance frbm  local mlarket in hours (logm lcal mrketsnchous (lo
Agricultural wage employment                    Agricultural self-employment
Figure 2. Employment and distance from local market



1.93 -                                      0.63
o                                           0
o 
-0.35 _____________________________         -0.12 ____________________________
distance fiom town In hours                 distance t4om town in hours
Non-farm wage employment                    Non-farm self-employment
0.35 -                                      0.33 -
0                                           0
o 
o                               ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~o
-0.60 __________________________            -1.93 __________________________
distance frm town in hours                  dsac4rntown in holurs
Agricultural wage employment                Agricultural self-employment
Figure 3. Employment and distance from towns



0.07                                 0.14
D0                                   0
E                                    E
0                                    0
*0 
0                                    0
o                                    0
DiFgstancefrom loal  ma re t            distancef wnmnhons
k.n
Figure 4. W'or'Kmigration and distance



0.74                                             4.27
o                                                c
0
o~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~c
0.00                                             0.00
.12                  2                                         makti1hus(og2ae
2 0stance frm local market in hours (log scale)  b2stance fm locl makt in hours (logwale$
Size of operated land                              Livestock sales
2.96
0,
0.00 
1i2        .5l1                       lb
.t2 lrstance om local market In hours (log scale)
Fertilizer per hectare
Figure 5. Farming and distance from local market



0.13-                                         4.37 -
0)~~~~~~~~~~~~~~~
-1 .71 -/-1 2.34-/
1       ~~~~distance fEom town in hours  1b1distance ftom town in hoturs
Size of operated land                            Livestock sales
14.59-
0~~~~~~~~~~~~~~
.G
w
-0.79 -_
-1.71  Size distance IAom town in hours
Fertilizer per hectare
Figure 6. Farming and distance from                sowns



0.55                                        0.24
0 t                                         0 .3
co                                         I-
-~~~~~~~                          Cu~~~~~~~0
o                                -~~~~~~~~~~~~~~~~~~~~~~~~~~~t
0.00                                        0.00 
.t   3stance Arm local market in hours (1log scalei  1b12  4111tanoe from local market in hours (log scaele
Wet Season -pulses                        Wet Season-vegetables
C      F        7i
0.00 _________________________              0.00 ________________________
frixn~~~~~~~~~~~~~~~~~pstne4 local market in hours. (log scale)  akt  uso
Wet Season - paddys                       Wet Season -- vegetables
C-fFgrl.Crpigpten                             n      itnefrmlclm              re



0.86 -0.63-
0~~~~~~~~~~~~~~~~
0                                 -~~~~~~~~~~~~i 
-                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~CD
distance ~iA  j~         t                    distance fLontown in0~r
Wet season - paddy                          Wet season - vegetables
5>-   ~0.22 -0.71-
06                                            0
122
Cu                               ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~U)
C~~~~~~~~~~~~~~~~~~~r                 C                      3r oni or
4-                                        4-~r eso  use  r  eao     egtbe
0                                          0iue8  Copn   atenaddsanefo   o n



0.17 -                                           0.68 -
o                                                0
o                                                0
0.00                                             -0.17
.12   *t       oa                                  ~moni or
.t2 stance fom local market in hours (log scale1       distance 4om town in hours
Distance from local market                          Distance from towns
Ln
LF
Figure 9. Crop Sales and distance



share of consumpbon  share of consumption
-n 
CD
-.                   share of consumption  share of consumpUon
3~~~~~~~~~~~~~~~~~~~~~~~
0
cnsar                        share of consurpo   h n) fxw pbon
0~~~~~~~~~~~
C~~~~~~~~~~~~~~C
!X ~~~~~~~~rt



0.80 -1.82-
o                                 ~~~~~~~~~~~~~~0
E                                              E
o                                              0
0.22 ___________________________              -0.23 - ________________________
C11stnc fbmloalmarket In hours (log scale,       distance &Mtowni  or
Distance from local market                        Distance from towns
Figure 1 1. Crop purchases and distance



0.85                                         0.16 -
-a                                           _
00
(A(
2L                                           C,
.0                                           0
.12  '35stance om  local ma rket in hors (log s.al 1    distance r  town in hours
Distance from local market                      Distance from towns
LnF22035.23 -                                 6029.36 -
00
IDI
-11235.06 --47954.89_________________________
.12   2tner oa              cl~     01twihus1
(41tnc   fbmloalm'arket in hours (log sc ,e  ldistance  towriamur
Distance from local market                      Distance from towns
Figure 12. Market surplus and distance



3 81 -                                       ¶76                                          0.29-
b                                            0<X * ;0               0 
E
E=                                           E
3  6  2  11.tanm                8    lm 121 5t 1$                     I      8b 1 48  2.25             8     1    2       82  lb,
d16mle mm  ocl  olm  I 5cm(kog scale)    dm000 m  local MI,.1kt1. h0cm (log scl)allna in    loa n  . hOInoM (logscle
Age                                     Household size                                 Adult males
0 19                                         ¶ 102 -j                                     063
° °°                                       0  00                                         -0.02
f2    8                            lb        2    .26                           i b      .12   Isk 8      M1.0 In 2  po   8$e   lb
Idaltn   (mm loal mdLt 1I tlm (log scale)  ld3ionce f   local mrout I6 loom (log ecab)  lboanCce hor local maak  I in5h  (log saee)  lb
Children                              Education of adult males                    Education of adult females
Figure 13. Household characteristics and distance from local market



09
share of members  number (in log)
i,              r 0        0   tDI 
CD                             D
I'                a
CL
CD 
ci)
0               C
En~~~~~~~~~~~~~~~~~~~a
(D~~~~~~~~~~~~~~~D
0~~~~~~~~
(0
cin



0.68                                         0.67
0.00                                         0.02-/
t2 gistance from local market in hours (log scalel  .12 3lstance from local mlarkeet in hours (log scalel
Inherited wealth                             Inherited wealth
o                                            0
C                                            C
9.40 -                                \-0.38 -\
.12
.12 istance frbm local market in hours (log scale1  distance .rFo   town in ho2urs
Distance from local market                      Distance from towns
Figure   15. Inherited wealth, food expenditures, and distance



-B               -           294 -                       004B
E                                                        E1
s                   ~~~~~~~~~~~~~~~~~~~~02       e 03
-1324 -                                                     It?  _AM     In_ .1.5   -  . _ _
.2  S1t  jbij l  2 L h hL0 1b dl  6 zb e  dh12  Sb  b)  lb  d_  6g,   , 009 ni ) 1
dhMonin oums d09             A.  04m in g nowl
Rice                       Other crops               Livestock products
Figure 16. Consumption and distance with household characteristics






Policy Research Working Paper Series
Contact
Title                           Author                  Date              for paper
WPS2820 The Privatization of the Russian  Igor Artemiev          April 2002        V Joseph
Coal Industry: PoliclEs and Processes Michael Haney                       32155
in the Transformation of a Major
Industry
WPS2821 Income, Wealth, and Socialization  Daniel Lederman       April 2002        P. Soto
in Argentina: Provocative Responses                                       37892
from Individuals
WPS2822 An Econometric Analysis of the    David Mckenzie         April 2002        C. Mendoza
International Bank for Reconstruction                                     80599
and Development's Creditworthiness
WPS2823 Real Exchange Rate IJncertainty   Luis Serven            April 2002        P Soto
and Private Investment in Developing                                      37892
Countries
WPS2824 Trade Policy and Labor Services:  Maurice Schiff         April 2002        P. Flewitt
Final Status Options for the West                                         32724
Bank and Gaza
WPS2825 Demand for Imports in Venezuela:  Mario A. Cuevas        April 2002        M Geller
A Structural Time Series Approach                                         85155
WPS2826 Potential GDP Growth in Venezuela:  Mario A. Cuevas      April 2002        M Geller
A Structural Time Series Approach                                         85155
WPS2827 Learning to Export: Evidence from  Marcel Fafchamps      April 2002        E. Khine
Moroccan Manufacturing          Said El Hamine                            37471
Albert Zeufack
WPS2828 Beyond Oaxaca-BlindE!r: Accounting  Fran,ois Bourguignon  April 2002       F. Bourguignon
for Differences in Household Income  Francisco H. G. Ferreira             31056
Distributions across Countries  Phillippe G. Leite
WPS2829 The Spatial Distribution of Poverty  Nicholas Minot      April 2002        R. Bonfield
in Vietnam and the Potential for  Bob Baulch                              31248
Targeting
WPS2830 Labor Market Implications of      Lodovico Pizzati       April 2002        L. Pizzati
Switching the Currency Peg in a                                           32259
General Equilibrium Model for Lithuania
WPS2831 Health Outcomes in Poor Countries  Limin Wang            April 2002        L. Wang
and Policy Options: Empirical Findings                                    37596
from Demographic and Health Surveys
WPS2832 Poverty and Survival Prospects    Adam Wagstaff          April 2002        H. Sladovich
of Vietnamese Children under Doi Moi Nga Nguyet Nguyen                    37698
WPS2833 School Choice, Student Performance, Emiliana Vegas       April 2002        H. Sladovich
and Teacher and School Characteristics                                    37698
The Chilean Case
WPS2834 Investor Protection, Ownership, and  Charles P. Himmelberg  April 2002     K Labrie
the Cost of Capital             R. Glenn Hubbard                          31001
Inessa Love



Policy Research Working Paper Series
Contact
Title                            Author                  Date               for paper
WPS2835 A Decade of Fiscal Transition      Asad Alam              May 2002           A Panton
Mark Sundberg                              85433
WPS2836 Ethnic Minority Development in     Bob Baulch             May 2002           E. Khine
Vietnam: A Socioeconomic         Truong Thi Kim Chuyen                      37471
Perspective                      Dominique Haughton
Jonathan Haughton
WPS2837 Reform, Growth, and Poverty        David Dollar           May 2002           E. Khine
in Vietnam                                                                  37471
WPS2838 Economic Mobility in Vietnam       Paul Glewwe            May 2002           E. Khine
in the 1990s                     Phong Nguyen                               37471
WPS2839 Marketing Externalities and        M. Shahe Emran         May 2002           F. Shilpi
Market Development               Forhad Shilpi                              87476
WPS2840 Public Spending and Outcomes:     Andrew Sunil Rajkumar   May 2002           H. Sladovich
Does Governance Matter?          Vinaya Swaroop                             37698
WPS2841 Contractual Savings in Countries   Gregorio Impavido      May 2002           P. Braxton
with a Small Financial Sector    Alberto R. Musalem                         32720
Dimitri Vittas
WPS2842 Financial Sector Inefficiencies    Pierre-Richard Agenor  May 2002           M. Gosiengfiao
and the Debt Laffer Curve        Joshua Aizenman                            33363
WPS2843 A Practical Guide to Managing      David Scott            May 2002           L. Yeargin
Systeniic Financial Crises' A Review                                        81553
of Approaches Taken in Indonesia,
the Republic of Korea, and Thailand
WPS2844 Money Demand in Venezuei,          Mario A. Cuevas        May 2002           M Geller
Multiple Cycle Extraction in a                                              85155
Cointegration cramework