Directory UMM :Data Elmu:jurnal:UVW:World Development:Vol29.Issue3.2001:

World Development Vol. 29, No. 3, pp. 481±496, 2001
Ó 2001 Elsevier Science Ltd. All rights reserved
Printed in Great Britain
0305-750X/01/$ - see front matter

www.elsevier.com/locate/worlddev

PII: S0305-750X(00)00110-8

Intersectoral Transfer, Growth, and Inequality
in Rural Ecuador
CHRIS ELBERS
Vrije Universiteit Amsterdam, The Netherlands
and
PETER LANJOUW *
Vrije Universiteit Amsterdam, The Netherlands and The World Bank,
Washington, DC, USA
Summary. Ð In this paper we study intersectoral transfer and its impact on the distribution of
income in Ecuador. We ®nd that income shares between farm and nonfarm activities are roughly
equal, on average, although the rich in rural areas typically receive a greater share of income from
nonfarm sources. Thus decomposing inequality by income source reveals that a rise in nonfarm

incomes increases inequality. Drawing on a new method to estimate local-level distributional
outcomes, growth of the high-productivity nonfarm sector is observed to have a strong and positive
association with average consumption and inequality. Growth of the low-productivity nonfarm
sector is associated with little change in either average income or income inequality. Irrespective of
subsector, growth of the nonfarm sector is associated with a substantial fall in poverty. Ó 2001
Elsevier Science Ltd. All rights reserved.
Key words Ð nonfarm employment, poverty, inequality, Lewis model, Ecuador

1. INTRODUCTION
The interaction between economic growth
and the distribution of income has been a
central theme of economics since its inception.
In the sub®eld of development economics, an
early and in¯uential view of the development
process was set out by Lewis (1954), in which
growth takes place against a background of
labor transfer out of traditional subsistence
agriculture toward a modern sector, often
tacitly assumed to be industrial and urban.
Fields (1980, 2000) demonstrates that such a

process is able to generate the well-known
``Inverted U-Curve'' of rising and then falling
income inequality, ®rst described by Kuznets
(1955, 1963). 1
This paper revisits some of these classic
themes in the context of rural Ecuador and
poses two speci®c questions. First, must the
process of intersectoral transfer necessarily
occur between the rural and urban sectors, or
can one view the rural nonfarm sector as an
alternative to the modern urban sector descri481

bed by Lewis? Second, what are the distributional consequences in rural areas of a growing
nonfarm sector? 2 The paper sheds empirical
light on these questions on the basis of household survey and census data for Ecuador. 3
We illustrate in this paper that the nonfarm
sector in rural Ecuador is quite large, and very
diverse. Because big di€erences in productivity
are observed, with many of the poor involved
in low-productivity, residual activities, it is not

immediately obvious whether the nonfarm
sector is inequality-increasing or inequality-reducing. We show, however, that income shares
from nonfarm activities are in fact highest
among the rich. This suggests that the nonfarm
sector is, on balance, inequality-increasing.

* We are grateful to Hans Hoogeveen, Jenny Lanjouw,
Martin Ravallion, Mitch Renkow, and two anonymous
referees for useful comments and suggestions. The views
in this paper are those of the authors and should not be
interpreted as those of the World Bank or any of its
aliates. All errors are our own.

482

WORLD DEVELOPMENT

This is con®rmed when we decompose income
inequality by source, although the elasticity of
inequality with respect to nonfarm income in

rural areas is low.
We explore further the distributional implications of the notion that the nonfarm sector
comprises a low-productivity and a high-productivity subsector (as described, for example,
by Ranis & Stewart, 1993). 4 Building on a
recently developed technique for estimating
distributional outcome measures at geographically disaggregated levels, we investigate econometrically the relationship between the lowproductivity and high-productivity nonfarm
subsectors and economic welfare. Taking
Ecuador's ``villages'' (parroquias) as our unit of
analysis, we ®nd that village-level income
inequality rates tend to rise with the share of
the village labor force employed in high-productivity nonfarm activities. Employment in
low-productivity nonfarm activities has either
no, or a negative, correlation with inequality,
depending on geographic region. Alongside
their association with income inequality, we
®nd that employment shares in the high-productivity nonfarm sector are signi®cantly (and
positively) correlated with village average per
capita consumption levels, and negatively
correlated with poverty rates. Our data allow us
to assess to what extent the high-productivity

part of the nonfarm sector in Ecuador acts as
an engine-of-growth, contributes to income
inequality, and reduces poverty, thus performing a role originally ascribed by Lewis to the
modern sector. Similarly, we measure the
importance of the low-productivity nonfarm
sector in acting as a safety net to protect the
poor.
The layout of the paper is as follows. In the
next section we provide the empirical evidence
to document the size, variety, and importance
(in terms of employment shares) of the nonfarm
economy in rural Ecuador. Section 3 presents a
framework within which to view the relationship between the nonfarm sector and the
distribution of income. In Section 4, we analyze
household survey data to identify the individual, household, and community-level factors
which appear to in¯uence whether a person is
engaged in the nonfarm sector. We report on
our exercise of decomposing inequality in rural
Ecuador by income source. In Section 5, we
describe the construction of our census-based

dataset of village-level income inequality and
poverty outcomes, and present empirical evidence on the association of the nonfarm sector

with village average per capita consumption,
poverty and inequality. Section 6 concludes the
paper.

2. THE RURAL NONFARM ECONOMY
IN ECUADOR
A nationally representative household survey
®elded in Ecuador in 1995 (the Encuesta de
Condiciones de Vida, ECV) provides considerable detail on the extent and nature of nonfarm
activities undertaken in rural areas. The survey
covered a total of 5,760 rural and urban
households, and follows a multi-module format
based on the World Bank's Living Standards
Measurement Surveys (LSMS). We draw on
this survey to describe the basic features of the
nonfarm economy in rural Ecuador. 5
Table 1 presents a breakdown of nonfarm

wage employment shares by subsector of
employment and geographic region. 6 In all the
three regions of Ecuador, the proportion of the
working population employed in nonagricultural activities is substantial, ranging from just
over a quarter in the Oriente to more than 43%
Table 1. Nonfarm wage employment in rural ecuador
(principal and secondary occupations)
Percentage of
working population involved in
Fishing
Extraction
Manufacture
Textiles/Garments
Wood/Straw/
Leatherware
Utilities
Construction
Commerce
Restaurant/Hotel
Transport

Finance
Property/Management
Administration
Teaching
Social services
Community work
Domestic service
Total
a

Costa

Sierra

Oriente

0.0 (0.0) 0.0 (0.0)
8.0
(18.3)a
0.7 (1.6) 0.9 (2.4) 0.3 (1.1)

4.4 (10.1) 6.7 (17.9) 2.6 (9.2)
0.9 (2.1) 1.4 (3.7) 0.3 (1.1)
0.4 (0.9) 2.5 (6.7) 5.8 (20.6)
0.2 (0.5)
3.2 (7.3)
15.8
(36.2)
1.6 (3.7)
2.1 (4.8)
0.1 (0.2)
0.7 (1.6)

0.0 (0.0) 0.0 (0.0)
6.2 (16.6) 2.2 (7.8)
7.7 (20.6) 6.6 (23.4)
0.9
1.8
0.0
0.2


(2.4)
(4.8)
(0.0)
(0.5)

0.6
2.2
0.3
0.0

1.3
1.9
0.5
0.5
1.4

1.9
2.4
0.6
3.1

1.1

(5.1)
(6.4)
(1.6)
(8.3)
(2.9)

3.0 (10.6)
0.9 (3.2)
0.6 (2.1)
2.0 (7.1)
0.8 (2.8)

(3.0)
(4.3)
(1.1)
(1.1)
(3.2)

43.7
(100.0)

37.4
(100.0)

Column percentages provided in brackets.

(2.1)
(7.8)
(1.1)
(0.0)

28.2
(100.0)

ECUADOR

in the Costa (although this ®gure includes
®shing activities that are signi®cant in the Costa
but not elsewhere). Commerce activities are
particularly important in the Costa, while in the
Sierra manufacturing and construction are also
signi®cant.
Table 2 provides a breakdown of homebusiness activities and their contribution to
family employment in rural Ecuador. In total
just under half a million small ®rms were estimated to be operating in rural Ecuador in 1995,
providing employment to nearly 900,000
persons. 7 Most rural businesses are quite
small, with an average of 1.8 workers. More
than four-®fths of all persons employed in
home businesses are family members. On
average, more than two-thirds of all self-owned
businesses are home-based. The total range of
activities in which home businesses are engaged
is quite large, but more than 40% of them are
involved in small-scale commerce, such as
shops selling basic provisions, restaurants, etc.
Other important sectors include agricultural
goods and food processing (4% of businesses),

483

®shing (7%), textiles and garments (9%) wood
and straw crafts (4%), transport services (5%)
and other services.

3. NONFARM EMPLOYMENT AND
INCOME INEQUALITY: A FRAMEWORK
OF ANALYSIS
In this section we consider a highly simpli®ed
scheme of income generation within which to
assess the interaction of the nonfarm sector
with income inequality. The scheme is very
much in the spirit of the Lewis (1954) model of
development. We consider three types of
activities in rural areas, summarized in Table 3.
The third activity in Table 3 refers to a lastresort activity for those who can ®nd no other
employment, or income source. One should
think of residual activities, such as very simple
trade or irregular services, generating a very
low income wlow , probably well below the
poverty line. 8 In a well-functioning labor
market wlow should equal marginal productivity

Table 2. Nonfarm rural enterprises in Ecuadora
No. of
enterprises

No. of
workers

No. of family
workers

Percent
home-based (%)

Total
employment

Agriculture (sales/services)
Forestry
Fishing
Mining/Extraction
Food processing
Textiles and garments
Leather goods
Wood and straw crafts
Paper
Sound/Recording
Rubber goods
Metals
Metal products
Machinery and equipment
Automotive
Furniture
Construction
Sales/Repair of vehicles
Wholesale commerce
Petty commerce
Hotel/Restaurant
Transport services
Financial intermediation
Machinery rental
Administration/Managerial
Teaching
Other services

9,056
2,152
34,440
4,319
9,074
40,537
1,529
20,235
633
486
425
6,466
2,274
573
727
14,250
10,547
3,312
1,179
194,760
13,855
21,482
340
547
3,020
2,667
71,797

2.37
2.37
1.89
6.61
2.09
1.37
2.01
1.59
1.00
1.00
3.63
3.06
2.45
1.00
1.94
2.11
2.41
1.25
2.55
1.72
2.29
1.83
3.00
2.32
1.27
1.17
1.45

1.44
1.53
1.28
1.63
1.80
1.29
2.01
1.33
1.00
1.00
0.12
1.83
1.09
1.00
1.94
1.81
1.48
1.00
1.83
1.56
2.14
1.25
2.00
1.32
1.00
1.07
1.13

55
58
4
92
95
99
100
85
100
100
100
100
81
100
94
94
68
98
47
75
81
1
100
32
59
100
69

21,477
4,815
65,294
28,563
19,027
55,513
3,074
32,367
633
486
1,544
19,783
5,570
573
1,409
30,090
25,418
4,132
3,008
335,010
31,727
39,235
1,020
1,268
3,844
3,129
104,188

Total

470,682

1.79

1.44

69

842,197

a

Source: Encuesta de Condiciones de Vida (1995).

484

WORLD DEVELOPMENT
Table 3. Types of rural activities

1. Agricultural activities
Required inputs: labor and land
2. Rural nonagriculture, high-productivity activities
Required inputs: labor and speci®c opportunities
(capital, connections, etc.)
3. Rural nonagriculture, low-productivity activities
Required inputs: labor

of labor in agriculture. 9 Hence one would
expect the wage rate wlow to be close to the
incomes of landless laborers in agriculture. Of
course, the incomes of the owners of land and
other nonlabor production factors in agriculture (the ®rst activity listed in Table 3) can be
much higher, say wlow ‡ p, where p is the pro®t
per land owner.
To work in the high-productivity, nonagricultural sector (activity 2 in Table 3) one needs
to possess special skills or opportunities. These
might come from education or, if this segment
of the labor market is characterized by information problems, or is otherwise not perfectly
competitive, these might come from other
individual-level opportunities to ®nd employment in this sector (such as relatives, friends, or
bribed civil servants). Typically the workers in
this sector are much better o€ than the lowproductivity workers in and outside agriculture, so that wage rate whigh is much higher than
wlow and could well be higher than the income
of land owners.
Suppose that the process of economic development in a particular setting takes the form of
a Lewis-style enlargement of a modern sector
(for example, the expansion of an industrial
sector with accompanying services sector). But,
suppose that instead of occurring in some
distant urban setting, and accompanied by
migration from rural areas to the cities (as
described in the original Lewis story), this
process takes the form of an expansion of highproductivity nonagricultural employment. 10
An example of such a transition is presented in
Table 4.
Going from employment Pattern I to Pattern
II in Table 4, it is clear that the incidence of

poverty is lower (since fewer people are working for a below-poverty line wage), average
income is higher, and inequality could very
well have risen. 11
In this simple characterization of the development process, the expansion of nonagricultural, high-productivity activities raises
incomes, while it may lead to a higher degree of
measured inequality. Moreover (also very much
following the Lewis model) one would expect to
see indirect e€ects deriving from increased
employment in the high-productivity sector. As
low-productivity labor becomes relatively more
scarce, increasing its opportunity cost, one
would expect the low-productivity wage rate
wlow to go up between Patterns I and II. Thus
the growth of high-productivity employment
could eventually trickle down to increase labor
productivity in the low-productivity sectors as
well, thereby helping to reduce poverty.
Below we further examine the rural nonfarm
economy in Ecuador, with an eye toward
assessing how well the stylized scheme described above is re¯ected in the data. 12 Note,
however, that the statistical analysis that
follows stands on its own and is not meant to
be a formal test of the framework of analysis
described above.

4. NONFARM INCOMES, ACCESS, AND
DISTRIBUTION: EVIDENCE FROM
SURVEY DATA
(a) Income shares
Total income from nonagricultural activities
derives from wage employment and home
businesses. Table 5 indicates that for the rural
population in Ecuador more than 40% of
income derives from nonagricultural activities,
only marginally less than the farm-income
share. 13 The nonagricultural sector in Ecuador
is thus signi®cant not only in terms of
employment but also income. Across quintiles
(de®ned in terms of per capita consumption
expenditure), the share of total income from

Table 4. Hypothetical employment patterns
Employment

Wage
Nonagriculture, low-productivity
Agriculture, landless labor
Agriculture, land owners
Nonagriculture, high-productivity

wlow
wlow
wlow ‡ p
whigh

Pattern I (%)

Pattern II (%)

50
25
25
±

25
25
25
25

ECUADOR

485

Table 5. Sources of income by expenditure quintile in rural Ecuador share of income from the respective sourcesa
Farm
(%)

Nonfarm

Agricultural
labor (%)

Other
(%)

Enterprise
(%)

Labor
(%)

Total
(%)

Poorest quintile
2nd
3rd
4th
5th

69
46
46
41
27

6
13
14
8
6

16
26
28
37
52

6
11
9
9
12

22
37
37
46
64

3
4
3
5
3

Total

46

9

32

9

41

4

a

We distinguish the following sources of income: (i) income from cultivation (including also ®shing); (ii) wage labor
income in agriculture; (iii) income from self-employment and family-enterprise activities (iv) wage and salaried
income from nonagricultural sources; (v) income from other sources (transfers, property, remittances, etc.).

nonagricultural sources rises sharply with living
standards. But, even the poorest quintile in
rural Ecuador receives about one-®fth of its
total income from nonagricultural activities.
This rises to 37% for the second and third
quintiles, and is as high as 64% for the top
quintile.
Between the two types of nonagricultural
income sources, home-enterprise income is
consistently more important as a fraction of
total income than nonagricultural labor income.
Again, the correlation between share of income
from home enterprises and consumption rankings is marked. The nonagricultural wage
income share represents only about 9% of total
income on average. This source is also less
monotonically linked to consumption rankings
than home-enterprise income. While the poorest
quintile receives 6% of total income from
nonagricultural wage labor sources, this rises to
11% for the second quintile, falls back to 9% for
the next two quintiles, and then rises back to
12% for the top quintile.
(b) Employment probabilities and determinants
of wage-labor earnings
We now turn to the factors which are associated with employment in nonagricultural
activities, and the level of earnings that such
activities generate. Table 6 presents three
Probit models linking the probability of having
primary employment in a nonagricultural
wage-labor occupation to a range of explanatory variables. 14 In the ®rst regression, the
dependent variable takes a value of 1 if the
person is primarily employed in nonagricultural
wage labor and 0 otherwise. The second and
third models split those employed in the
nonagricultural wage-labor force into two

groups; those with a low-productivity job and
those with a high-productivity job, respectively.
The distinction between low- and high-productivity is based on whether earnings, respectively, fall below, or exceed, the average
earnings of someone with agricultural wage
labor as a primary occupation.
Considering all nonagricultural employment
together, women are signi®cantly more heavily
represented in the nonagricultural wage-labor
force than men. At average values of all other
variables, the probability of primary employment in the nonagricultural sector rises from
8% for a man to 21% for a woman. 15 What is
striking, however, is that after dividing the
types of occupations into two groups depending on whether earnings are on average lower
or higher than average earnings from agricultural labor, women are signi®cantly less likely
to be employed in the relatively high-productivity occupations. The likelihood of being
employed in a high-productivity job falls from
1.2% for a man to 0.6% for women, at average
values of all other variables.
Relative to the uneducated, those with education are generally more likely to be employed
in the nonagricultural sector, particularly in the
high-productivity jobs. In low-productivity
jobs, the only statistically signi®cant education
variable is a dummy variable for secondary
education. In the high-productivity jobs the
primary, secondary, and university education
dummies are all statistically signi®cant. At
average values of other variables, having
completed primary education raises the probability of employment in a high-productivity
job from 0.3% for the uneducated to 1%.
Education at the secondary-level increases this
probability to 5%. The probability of being
employed in a high-productivity job then jumps

486

WORLD DEVELOPMENT
Table 6. Probability of nonagricultural employment as a primary occupation
Probit model

Intercept
Household size
Female
Age
Age squared
Quichua speaker
Shuar speaker
Pre-primary education
Primary school education
Secondary school education
University education
Other tertiary education
Post-graduate education
Vocational training
Land owned per capita
Land owned Squared
Cultivating household
(dummy)
Rural periphery
Rural dispersed
Costa
Oriente
Migrant during past decade
Log likelihood (model)
Log likelihood (constant)
Total observations
Observations at 0
Observations > 0
LR test (model)
Degrees of freedom
Critical v2

All employment in
nonagricultural sectora

Employment in
low-productivity jobb

Employment in
high-productivity jobb

Estimate

Prob.

Estimate

Prob.

Estimate

Prob.

)1.674
)0.006
0.642
0.073
)6E)4
0.102
0.419
0.186
0.253
0.604
0.777
7.344
5.592
0.127
)0.018
2.3E)6
)1.026

0.0001
0.5428
0.0001
0.0001
0.0001
0.3076
0.2392
0.2929
0.0017
0.0001
0.0045
0.9986
0.9993
0.4244
0.0056
0.0394
0.0001

)1.551
)0.020
0.852
0.035
)2E)4
)0.007
0.061
0.248
0.053
0.307
)0.428
7.493
)5.720
0.118
)0.025
3.3E)6
)0.620

0.0001
0.0660
0.0001
0.0001
0.0013
0.9473
0.8950
0.1834
0.5311
0.0066
0.2268
0.9986
0.9993
0.4896
0.0030
0.0171
0.0001

)3.073
0.021
)0.248
0.101
)0.001
0.156
0.694
0.025
0.435
0.669
1.299
)5.070
6.722
0.003
)0.003
)2E)5
)0.939

0.0001
0.1111
0.0012
0.0001
0.0001
0.3296
0.0894
0.9253
0.0004
0.0001
0.0001
0.9994
0.9995
0.9894
0.7868
0.8860
0.0001

)0.784
)0.863
0.247
)0.357
0.033

0.0001
0.0001
0.0001
0.0002
0.6695

)0.416
)0.646
0.293
)0.323
)0.016

0.0006
0.0001
0.0001
0.0035
0.8497

)0.812
)0.536
)0.002
)0.156
0.036

0.0001
0.0001
0.9806
0.2160
0.7151

)1479
)2147

)1248
)1618

)815
)1109

4523
3699
824

4523
4001
522

4523
4221
302

1336
21
32.67

740
21
32.67

588
21
32.67

a
Nonagricultural employment here denotes only those individuals with wage employment in the nonagricultural
sector as a primary occupation.
b
Low-productivity and high-productivity jobs have been designated as such if the annual earnings derived from them
fall below or above, respectively, the average annual per capita income from agricultural wage labor for persons
engaged in agricultural wage labor as a primary occupation.

to 37% for individuals with education levels at
the university level. It is important to
acknowledge that the exogeneity of education
in these models can be questioned, so one must
be careful to refrain from concluding that
improvements in education would necessarily
lead to increased employment in high-productivity nonagricultural occupations. The evidence does suggest, however, that this question
merits further research.
In all models, age is positively associated
with the probability of nonagricultural
employment up to about 55 years of age in the
full model. Beyond that age, the probability of

nonagricultural employment declines. The
corresponding turning points in the low-productivity and high-productivity models are 65
and 50, respectively. Being indigenous plays a
role only in the case of Shuar speakers and the
high-productivity jobs.
Individuals from households which report
some income from cultivation are signi®cantly
less likely to be employed in the nonfarm sector
in all three modelsÐpresumably because for
cultivating households the ®rst call on family
labor is on the farm. Per capita land holdings
have a signi®cantly negative e€ect on nonagricultural employment and, given nonagricultural

ECUADOR

employment, on employment in low-productivity
jobs. The nonsigni®cance of this variable
(indeed, the zero estimate of the relevant coef®cient) in the high-productivity regression,
suggests that a positive e€ect is neutralizing this
general negative e€ect. It is sometimes argued
that if o€-farm employment opportunities,
particularly the more attractive ones, are
rationed, then access might be in¯uenced by the
household's wealth and in¯uenceÐand this
might be correlated with landholding (see
Lanjouw & Stern, 1998). This would imply a
positive and signi®cant coecient on land in
the case of the high-productivity jobs. While
the evidence here does not o€er strong support
to this contention, the lack of a strong negative
relationship between landholding and highproductivity nonfarm jobs suggests that one
cannot exclude a wealth e€ect.
The ECV95 data disaggregate rural areas into
three subregions; rural periferia, rural amanzanado, and rural disperso. Rural periferia refers
to the rural areas immediately surrounding the
larger conurbations. Rural amanzanado corresponds to rural communities with some basic
infrastructure but a population of less than
5,000 persons. Rural disperso refers to the
remaining rural areas. In Table 6 it appears
that, relative to persons living in the rural
amanzanado areas, persons from both the urban
periphery and outlying areas are less likely to be
employed in the nonagricultural sector (both
high- and low-productivity jobs). In the case of
the outlying areas this is not surprising, as
presumably households are more likely to be
engaged in cultivation there. But, the lower
probability of nonagricultural employment for
persons in the urban periphery is puzzling: one
would think that wage employment opportunities are relatively common in urban centers.
However, poverty in the urban periphery is
much higher than in either the amanzanado areas or in urban areas (Lanjouw, 1999).
Several factors might combine to explain this
observation. First, periurban areas might function as a temporary stepping stone for migrants
from remote rural areas aiming to enter the
urban sector. As such, few are likely to be
prepared to make the investments necessary to
establish substantial nonfarm activities. Second,
proximity to large urban markets might prompt
intensive agricultural activitiesÐparticularly
the cultivation of perishable foodcrops which
can be sold on the urban markets.
Relative to those living in the Sierra, the
population in the Costa is more likely to be

487

employed in nonagricultural activities. This is
not, however, signi®cant in the high-productivity job model, suggesting that while there
may be more nonagricultural activity in the
Costa, much of this is relatively low paid. This
observation is consistent with the ®nding in
World Bank (1995) that in the Costa the poor
are more widely engaged in both the agricultural and nonagricultural labor markets, while
the poor in the Sierra are often subsistence
cultivators. In the Oriente the probability of
nonagricultural employment is lower than in
the Sierra, particularly in the low-productivity
jobs.
In Table 7 we examine earnings from
nonagricultural jobs on the basis of an OLS
regression for the subset of persons with

Table 7. Nonagricultural wage labor incomea
OLS modelb
Intercept
Household size
Female
Age
Age squared
Quichua speaker
Shuar speaker
Pre-primary education
Primary school
education
Secondary school
education
University education
Other tertiary
education
Post-graduate
education
Vocational training
Land owned per capita
Land squared
Cultivating household
(dummy)
Costa
Oriente
Migrant during past
decade
Mills ratio
Adjusted R2 : 0.267
Number of observations: 825
a

Estimate

Prob.
value

12.40
0.05
)1.22
0.106
)0.001
)0.03
0.54
0.03
0.27

0.0001
0.0028
0.0001
0.0001
0.0001
0.8853
0.4498
0.9248
0.0638

0.39

0.0382

1.27
)0.71

0.0638
0.7443

1.56

0.5011

)0.23
0.05
)5E)4
)0.47

0.3482
0.1586
0.6561
0.0001

)0.25
)0.15
0.10

0.0163
0.4359
0.4580

1.0E)8

0.8892

Nonfarm incomes are calculated as earnings from
individuals' primary wage employment. Household-enterprise incomes are therefore not included. Incomes are
expressed in annual sucres (in 1995 US$1.00 was
approximately equal to 3,000).
b
Dependent variable: (log) annual nonagricultural wage
income with adjustment for sample selection.

488

WORLD DEVELOPMENT

primary employment in the nonagricultural
sector. The speci®cation for this model includes
a correction for sample-selection based on the
®rst Probit model in Table 6. 16 The insigni®cant parameter estimate on the Mills ratio
variable suggests that in the present example
there is no correlation between unobserved
variables which in¯uence the probability of
employment in the nonfarm sector and unobserved variables a€ecting earnings in that
sector.
As we would expect, given the di€erent
probabilities of employment in low- versus
high-productivity jobs observed in Table 6,
women earn less than men from nonagricultural jobs. Based on the parameter estimates of
Table 7, a woman would expect to earn about
70% less than a man from her nonagricultural
occupation. 17 The association between
nonfarm earnings and education is very strong
(although once again, direction of causation
has not been established here).
In the Probit analysis of Table 6, there was
at best a weak suggestion that persons with
greater wealth (proxied by per capita land
holdings) might be more highly represented in
high-productivity nonagricultural occupations.
In terms of earnings this conjecture receives
additional support; an additional hectare
owned is associated with higher nonagricultural earnings (by about 5%). Once again,
however, if the household is cultivating some
land (as opposed to simply owning land),
earnings decline. A person from a cultivating
household earns about 37% less than if the
household is not cultivating. The probable
explanation for this is that a person who
belongs to a cultivating household may spend
at least some time helping on the farm (for
example, for harvesting and at other periods
of peak labor demand) and this reduces his
average monthly income from nonfarm
employment, even if the latter is his primary
occupation.
As suggested from the Probit models, while
those in the Costa were more likely to be
employed in the nonagricultural sector than
those in the Sierra, they earn signi®cantly less
from such occupations. A person with a
primary occupation in nonagricultural wage
employment in the Costa would earn about
22% less than a person in the Sierra. There is no
signi®cant earnings di€erential between the
Sierra and Oriente, although the point estimates also suggest that the di€erential would
favor the Sierra.

(c) Home-enterprise activities
Table 8 returns to a Probit model examination of the likelihood that a household will
possess a home enterprise. This model is at the
household rather than individual level, and
although roughly similar explanatory variables
are applied as in the previous models, a few
variables relating to infrastructure access were
added.
Education is again strongly correlated. If the
most educated family member has a primary
school education or a secondary school education, then the household is more likely to
own a business than a household where nobody
is educated. Further, those households in which

Table 8. Probability of rural enterprise
Probit model

Estimate

Prob. value

Intercept
Household size
Quichua speaker
Shuar speaker

)0.50
0.05
0.16
)0.12

0.0003
0.0001
0.1080
0.7295

Education of best-educated
household member
Pre-primary schooling
Primary school
Secondary school
University education
Other tertiary education
Post-graduate education

)0.05
0.19
0.20
0.21
)0.13
6.53

0.6478
0.0757
0.0009
0.1016
0.5721
0.9987

0.17

0.0106

All family members literate
Land owned by household
Cultivating household
(dummy)
Rural periphery
Rural dispersed
Costa
Oriente
Migrant during past
decade
Connection to electricity
network
Telephone connection
Water connection
Log likelihood (M): )1487.08
Log likelihood (0):
)1673.51
Total Observations: 2492
Observations at 0: 1504
Observations > 0: 988
LR test (model): 373
Degrees of freedom: 20
Critical v2 : 31.41

)0.00007
)0.30

0.7931
0.0001

)0.39
)0.62
0.15
0.15
)0.11

0.0014
0.0001
0.0136
0.1163
0.1037

0.26

0.0002

0.30
0.05

0.0744
0.4661

ECUADOR

all family members are literate are more likely
to own a business than those where nobody is
educated. Unlike in the case of employment,
however, higher levels of education appear to
be relatively less strongly correlated with
household business activity. This ®nding could
indicate that those with tertiary levels of education are more likely to enter into a salaried
occupation than set up a family business.
As before, cultivating households are less
likely to have a home business, and land
holding exercises no signi®cant independent
in¯uence. Those residing in the rural periphery
and outlying areas are once again less likely to
own family businesses. As before, the Costa
region has a relatively higher incidence of
family businesses than the Sierra.
Whether a household is connected to the
public electricity network, and whether it has a
telephone connection are strongly related to the
likelihood of home-enterprise ownership. These
observations add to the perception that infrastructure is an important facilitator of nonagricultural activity.
(d) Decomposing inequality by factor source
In this section we decompose income
inequality by factor components to assess the
contribution of sources of income to total
income inequality. What we can focus on here
is the elasticity of overall inequality, the degree
to which overall inequality changes with small
changes in rural nonagricultural incomes.
Table 9 presents the Gini coecient for
income in Ecuador as a whole (including urban
areas), decomposed by income components. 18
Following Shorrocks (1982) the Gini coecient
G, can be obtained as a weighted average of
``pseudo-Gini'' coecients Gi for each component, where the weights are given by the share
ak of component income in total income: 19

489

G ˆ a1 G1 ‡    ‡ ak Gk ‡    ‡ an Gn :
The pseudo-Gini coecient for an income
component is similar to the Gini coecient for
that component but with the modi®cation that
individuals are ranked in terms of their total
income rather than component income. 20
In general, the change in the overall income
inequality brought about by an increase or a
reduction of income from a given source will be
smaller the closer the pseudo-Gini coecient
for that source is to the overall Gini. To see
this, suppose we decompose income into two
components:
G ˆ aG1 ‡ …1

a†G2 :

Consider a change in income from a di€erent
mix of the two income sources, assuming that
the distribution between income sources does
not change
G0 ˆ a0 G1 ‡ …1

a0 †G2 :

This implies that
DG ˆ G0



Da…G2

G1 †:

Because
G2

G1 ˆ

…G
…1

G1 †
;


the change in G can be written as


Da
…G G1 †:
DG ˆ
1 a
The smaller the di€erence between the pseudoGini coecient for a given source and the
overall Gini coecient, the smaller will be the
impact on inequality from a change in income
from that source. The elasticity of the Gini
coecient with respect to a change in income
from component 1 is thus proportional to the
percentage di€erence between the overall Gini
coecient and this pseudo-Gini coecient

Table 9. Income inequality by factor components
Per capita incomes in rural Ecuador
Pseudo-Gini coecient (G)
Share of total per capita income (ak )
Gini coecient (Gk )
Coecient of rank correlation (Rk  Gk =Gk )
Contribution to overall Gini coecienta
Elasticity of overall Gini to small increase in
component income
a

Farm
income
0.791
0.372
0.926
0.854
38%
0.005

Agricultural
labor income
0.665
0.089
0.889
0.748
8%
)0.015

Rural nonfram
income

Other
income

Total
income

0.817
0.497
0.895
0.913
52%
0.040

0.611
0.042
1.055
0.579
3%
)0.01

0.785
1.000
0.785
1.000
100%

This can be calculated as the product of the corresponding entries in (1) the ®rst two rows, or (2) the second, third
and fourth rows.

490

eG
1 ˆ

WORLD DEVELOPMENT

DG
G



Da
ˆ
a



a
1

a

 G

G1
G



:

From Table 9 we see that the contribution of
nonagricultural income inequality to overall
inequality in rural Ecuador is 52%, compared
to 38% for farm income. With an elasticity of
0.04 we can see that an increase in rural
nonagricultural income increases overall
inequality in rural areas. An increase in rural
nonagricultural income of 10% would raise the
Gini coecient for rural Ecuador from 0.785 to
0.789.
It thus appears that on the basis of survey
data for rural areas, nonagricultural incomes
go primarily to the better o€, so that higher
nonagricultural incomes (as opposed to more
nonagricultural income earners) is associated
with higher inequality.

5. INTERSECTORAL TRANSFER,
GROWTH, POVERTY, AND INEQUALITY
A single-period household survey o€ers
limited scope to investigate directly the relationship between growth of the nonfarm sector
and average incomes and their distribution.
The decomposition exercise carried out in the
preceding section, for example, while suggestive, does not capture the indirect e€ects that
rising nonfarm incomes can exert on total
incomes and their distribution. In addition, the
exercise is at best able to point to the impact of
a small change in nonfarm incomes.
The detailed time-series data necessary to
investigate these dynamic questions more
completely are unfortunately not available for
Ecuador (nor for most other developing countries). In this section we describe an alternative
approach to investigate the impact of a growing
nonfarm sector. We scrutinize the relationship
between distributional outcomes and employment shares in the nonfarm sector at the level
of Ecuador's 1,000-odd rural parroquias, the
small administrative-delineated geographic unit
in the country. This is possible as a result of
recent research in which data from the 1994
ECV survey are combined with the 1990
Ecuadorean census (Elbers, Lanjouw, &
Lanjouw, 2000). The technique is brie¯y
described below.
Elbers et al. (2000) use data from the 1994
ECV survey to estimate a model of per capita
consumption expenditure and then use the
resulting parameter estimates to weight census-

based characteristics of the entire population of
Ecuador and calculate each household's
expected welfare level. They show that this
merging of data sources yields an estimator
which can be clearly interpreted, extended in a
consistent way to any aggregated welfare
measure (poverty rate, measure of inequality,
etc.) and which can be assessed for statistical
reliability. They show that the method yields
estimates of poverty and inequality measures
which are quite precise and reliable for populations of 5,000 households and still remarkably good for populations as small as 500
households. 21
Based on the method described in Elbers et
al. (2000), a database was constructed
comprising estimates of headcount rates, average per capita consumption levels, and
consumption inequality (calculated on the basis
of the Atkinson measure with an inequality
aversion parameter value of two) for each of
the 915 parroquias in rural Ecuador. This
database was combined with census-based
information on parroquia-level demographic
composition and employment shares in di€erent sectors. The resulting parroquia-level dataset constructed in this way permits one to carry
out an analysis within Ecuador along lines
similar to what has traditionally been carried
out at a cross-country level. 22
We estimate three sets of regression models,
with respectively the parroquia average per
capita consumption level, the headcount rate,
and the Atkinson 2 measure of inequality as
dependent variables. These dependent variables
are regressed on the same speci®cation
comprising the share of the working population
employed in low-productivity nonfarm activities, the share employed in high-productivity
nonfarm activities, and a set of variables
capturing the demographic composition of
each parroquia. 23 The models are estimated
separately for the three main agro-climatic
regions in Ecuador. The results are reported in
Table 10.
Considering ®rst the relationship between
nonfarm employment shares and per capita
consumption, we can see that in all three
regions, per capita consumption is considerably
higher when the share of the working population employed in high-productivity nonfarm
jobs is higher. In 1994 US dollar terms, an
additional percent of the working population
employed in the high-productivity nonfarm
sector is associated with an average per capita
consumption per month US$74 higher in the

ECUADOR

491

Table 10. Diversi®cation into the rural nonfarm sector and welfare: explaining parroquia level per capita consumption,
poverty and inequality
Dependent
variable

Average per capita consumption

Poverty
(headcount)

Inequality
(Atkinson 2 measure)

Parroquia-level
explanatory
variables

Sierra

Costa

Oriente

Sierra

Costa

Oriente

Sierra

Costa

Oriente

Share of labor
force in lowproductivity
nonfarm job
Share of labor
force in highproductivity
nonfarm job
% population
0±10 years
% population
10±20 years
% population
40±60 years
% population
60+ years
No. of
households
(thousands)
Constant

9.91
(3.88)a

)10.67
()1.55)

25.32
(2.57)

)0.13
()6.43)

0.08
(2.69)

)0.25
()2.28)

)0.02
()2.32)

0.02
(0.71)

0.04
(0.76)

74.11
(16.80)

82.23
(10.23)

58.55
(9.41)

)0.42
()12.10)

)0.41
(11.29)

)0.47
()6.94)

0.26
(13.89)

0.21
(7.50)

0.24
(6.63)

)180.68
()9.65)
)55.56
()2.47)
)149.37
()5.35)
)4.06
()0.18)
1.98
(6.63)

)238.02
()7.38)
)241.14
()5.68)
)15.77
()0.25)
)239.66
()4.00)
6.00
(0.76)

)76.96
()2.97)
)5.99
()0.20)
)27.10
()0.77)
120.84
(2.47)
9.22
(0.78)

1.80
(12.10)
0.12
(0.65)
1.62
(7.32)
)0.39
()2.16)
)0.009
()3.59)

1.68
(11.44)
0.82
(4.23)
0.26
(0.89)
0.70
(2.57)
0.005
(1.50)

0.67
(2.37)
)0.03
()0.09)
0.14
(0.38)
)1.61
()3.01)
)0.017
()1.33)

)0.20
()2.57)
)0.15
()1.58)
0.18
(1.56)
)0.43
()4.43)
0.001
(1.08)

)0.24
()2.11)
)0.38
()2.61)
0.01
(0.04)
)0.49
()2.37)
0.013
(4.61)

)0.54
()3.61)
0.20
(1.19)
)0.50
()2.46)
)0.36
()1.23)
)0.003
()0.48)

124.80
(9.43)
0.778
490

179.54
(7.73)
0.563
271

57.73
(3.27)
0.669
145

)0.07
()0.69)
0.750
490

)0.11
()1.07)
0.647
271

0.63
(3.30)
0.585
154

0.45
(8.10)
0.537
490

0.49
(6.14)
0.415
271

0.58
(5.66)
0.445
145

Adjusted R2
No. of
observations
a

t-statistics in brackets.

Costa region, US$82 higher in the Sierra region
and US$59 higher in the Oriente. More lowproductivity nonfarm employment is associated
with higher average consumption in the Sierra
and Oriente (albeit modestly), but with lower
average consumption in the Costa (although
this is not statistically signi®cant). In this latter
region, agricultural wage labor is widespread
and relatively well-paid in at least some places
(due to the location of many export-oriented
plantations in this region), so that low-productivity nonfarm activities are particularly
likely to indicate distress in the rural Costa
region. In the Sierra, with its widespread rural
manufacturing tradition, the low-productivity
category is less clearly a category of last-resort,
distress activities. In all regions, the coecients
on employment shares are statistically signi®cant (except for rural Costa with the low-productivity category). The parameter estimates on
the demographic variables indicate that average
consumption levels are highest in parroquias
with large working age populations. The total
population of the parroquia does not appear to

markedly in¯uence average consumption levels
(only in the Sierra is this parameter estimate
signi®cant, and even then it is small). This
simple speci®cation explains a remarkably high
78% of total variation of average per capita
consumption across parroquias in rural Sierra,
56% in Costa and 67% in Oriente.
The positive association of nonfarm
employment shares with average income is
mirrored by a largely negative correlation with
poverty. Once again, the share of the working
population in high-productivity nonfarm
employment has the most marked impact. An
additional percentage of the working population employed in high-productivity activities is
associated with a decline in the headcount of
42% points in the Costa, 41 in the Sierra and 47
in the Oriente. While more low-productivity
nonfarm activity is associated with lower
poverty in the Sierra and Oriente, the opposite
is true in the Costa. The response of poverty to
changes in the low-productivity employment
shares is milder than in the case of high-productivity nonfarm activities, but the correlation

492

WORLD DEVELOPMENT

with the low-productivity shares is far from
negligible. In the Sierra and Oriente the e€ect of
an increase in the low-productivity employment
share is between a third and half as strong as
the e€ect of a change in the high-productivity
employment share (compared to a relative
impact on average consumption of less than
one-eighth). A 1% increase in the share of the
working population employed in low-productivity nonfarm activities is associated with
poverty, respectively, 13% and 25% points
lower. In the Costa the association is with
poverty eight percentage points higher. It is
clear that the low-productivity category in the
Costa region is most clearly associated with
distress, while in the other two regions,
employment in low-productivity activities
remains more attractive than the default activities in agriculture. Once again a fairly high
degree of explanatory power is obtained in all
three regions (adjusted R2 's ranging between
0.75 and 0.59).
While a bigger nonfarm sector is clearly
associated with higher average incomes and
lower poverty, it is also associated with greater
consumption inequality. This is particularly the
case with the high-productivity nonfarm activities that were also associated most strongly
with higher consumption and lower poverty.
The low-productivity nonfarm activities are
either associated with lower inequality (Sierra)
or with no discernable impact on consumption
inequality. While the explanatory power of this
model is lower than in the other two models, it
remains quite high.
Overall, the evidence suggests that highproductivity nonfarm jobs raise average
incomes and result in higher income inequality. But they do also have a dampening e€ect
on poverty. It is not obvious that an expansion of high-productivity nonfarm activities
results directly in improved employment
opportunities for the poor. Yet, the impact we
observe on rural poverty is very strong. One
possibility is that the mechanism works
through the agricultural labor market: greater
employment in the high-productivity nonfarm
sector tightens the agricultural labor market
(as those farmers who are relatively well educated are drawn to work in the nonfarm
sector), and this results in greater participation
and/or higher wage rates by the poor in the
agricultural labor market. Other explanations
might also apply. It is important to emphasize
that the models estimated here have not been
able to establish the direction of causation

from the nonfarm sector to the welfare
outcomes. It is possible, for example, that
exogenously driven growth in agriculture
raises incomes and income inequality, and
reduces poverty, and that these changes in
well-being result in increased demand for
nonagricultural goods and services (so that
employment rates in these subsectors change).
The wide variety and strength of ``linkages''
between the nonfarm sector and the agricultural sector has been the subject of much
theoretical and empirical analysis. 24
6. CONCLUSION
Traditional theories of development paid
considerable attention to the role of intersectoral transfer (from agriculture to industry) as a
central feature of the development process.
There has also been a long-standing interest in
understanding how economic development
a€ects the distribution of income.
In this paper we have concentrated on the
case of a single developing country, Ecuador, to
study both the process of intersectoral transfer
and its impact on the distribution of income.
Our analysis shows that while traditional
theories tended to locate the modern sector in
urban areas, the rural nonfarm sector in
Ecuador is both large and extremely heterogeneous. Income shares in rural areas from
nonagricultural activities are only slightly
lower, on average, than from farming, with on
balance, the share of nonfarm income being
highest among the top quintile in the distribution of income. When we split nonfarm activities into two categories, low-productivity and
high-productivity activities, we show that the
pro®le of persons involved in these two
subsectors are quite di€erent. Women are
highly represented in low-productivity activities. The well educated are highly represented in
high-productivity occupations. There is evidence that nonfarm activities are most common
in and around rural townships, and when
households have better access to infrastructure
services.
We have shown, on the basis of a decomposition of inequality by income source, that a
rise in nonfarm incomes would increase
inequality. We pursue this question further at
the level of Ecuador's parroquias. Drawing on
a new method to estimate local-level poverty,
inequality and average consumption levels, we
show that growth of the high-productivity

ECUADOR

nonfarm sector has a strong and positive
association with average consumption and
also on inequality. Growth of the low-productivity nonfarm sector is associated with
little change in either average income or
income inequality. Growth in either subsector
of the nonfarm sector, however, is associated
with a substantial and negative impact on
poverty (with the single exception occurring in
the case of low-productivity employment in
rural Costa). We suggest the possibility that
expansion of high-productivity nonfarm
activities in¯uences poverty via a tightening of
the agricultural labor market that raises
participation rates and/or wage rates in the
agricultural labor market.
Overall our analysis suggests that the traditional Lewis model of growth through intersectoral transfer remains highly relevant once it
is recognized that the modern, nonagricultural
sector can develop in rural areas as well as in

493

the cities. Our evidence suggests that this
process is likely to be inequality-increasing, but
that this should not be interpreted to imply that
the poor do not bene®t.
A ®nal word on the limitations of our
analysis is warranted. We have been careful in
this paper to refrain from attributing a strong
causal link to the relationship between the
nonfarm sector and welfare outcomes. While
we believe that such a link may well apply,
our statistical analysis does not establish it. In
particular, we cannot exclude the possibility
that it is the agricultural sector that acts as
the driving force behind both changes in
welfare as we