Government Spending, Growth and
Poverty in Rural India
Shenggen Fan, Peter Hazell, and Sukhadeo Thorat
Using state-level data for 1970–93, a simultaneous equation model was developed to estimate the
direct and indirect effects of different types of government expenditure on rural poverty and productivity growth in India. The results show that in order to reduce rural poverty, the Indian government should give highest priority to additional investments in rural roads and agricultural research.
These types of investment not only have much larger poverty impacts per rupee spent than any
other government investment, but also generate higher productivity growth. Apart from government
spending on education, which has the third largest marginal impact on rural poverty and productivity growth, other investments (including irrigation, soil and water conservation, health, and rural
and community development) have only modest impacts on growth and poverty per additional
rupee spent.
Key words: public policy, government expenditure, rural poverty, productivity growth, India.
Poverty in rural India has declined substantially in recent decades. The percentage of the
rural population living below the poverty line
fluctuated between 50 and 65% prior to the
mid-1960s, but then declined steadily to about
one-third of the rural population by the early
1990s (table 1). Agricultural growth and food
price changes have been identified as important contributing factors to this decline in
rural poverty ( Saith; Ahluwalia; Srinivasan;
Ghose; Gaiha; Bell and Rich). But these factors may have become less important today
given growth in the rural non-farm economy
and expansion of government poverty alleviation and employment programs.
The primary purpose of this paper is to
investigate the causes of the decline in rural
poverty in India and particularly to determine the role that government investments
have played, using a pooled time-series, crossstate data set. Government spending can
have direct and indirect effects on poverty.
The direct effects are the benefits the poor
receive from expenditures on employment
programs. The indirect effects arise when
government investments in rural infrastructure, agricultural research, and the health
Shenggen Fan and Peter Hazell are senior research fellow and
director, respectively, of the Environment and Production Technology Division, International Food Policy Research Institute,
Washington, D.C. Sukhadeo Thorat is professor at Jawaharlal
Nehru University, New Delhi. They thank two anonymous
reviewers, workshop participants at IFPRI, and World Bank for
comments.
and education of rural people stimulate agricultural and nonagricultural growth, leading
to greater employment and income-earning
opportunities for the poor and to cheaper
food. We seek to quantify the effectiveness of
different types of government expenditures
in contributing to poverty alleviation. Such
information can assist policymakers in targeting their investments more effectively to
reduce poverty. More efficient targeting has
become increasingly important in an era of
macroeconomic reforms in which the government is under pressure to reduce its total
budget. An econometric model is formulated and estimated that permits calculation
of the number of poor people raised above
the poverty line for each additional million
rupees spent on different expenditure items.
Targeting government expenditures simply
to reduce poverty, however, is not sufficient.
These expenditures also need to stimulate economic growth to help generate the
resources required for future government
expenditures. Growth is the only sure way
of providing a permanent solution to the
poverty problem and increasing the overall welfare of rural people. Our model is
therefore formulated to measure the impact
on growth as well as poverty of different
items of government expenditures. The model
makes it possible not only to rank different
types of investment in terms of their effects
on growth and poverty, but also to quantify
Amer. J. Agr. Econ. 82(4) (November 2000): 1038–1051
Copyright 2000 American Agricultural Economics Association
Fan, Hazell, and Thorat
Table 1.
Changes in Technology, Infrastructure, Agricultural Production and Poverty in Rural India, 1970–94
HYVs
Irrigation
Village
Electrified
Literacy
Rate
Road
Density
km/1000 km2
%
%
%
23
28
33
33
34
57
85
89
23
30
35
39
2414
3670
4861
5221
Standard Deviation
1970
15
1980
19
1990
23
1994
20
18
20
21
23
22
22
13
10
6
7
8
9
1563
2189
3242
3581
Annual Growth Rate (%)
1970–79
625
1980–89
310
1990–94
210
1970–94
419
184
169
−088
120
537
438
096
396
247
161
268
217
450
305
134
311
%
100
120
152
165
n.a.
27
16
18
195
382
209
211
Productivity
Growth
%
Incidence of
Poverty
%
10000
11205
12069
11786
55
48
36
37
n.a.
24
21
21
14
13
10
10
137
199
−059
069
Note: Standard deviations for production and productivity growth are calculated using the growth index between the current and last periods ( 1970–80, 1980–90, and 1990–94).
Government Spending, Growth and Poverty in Rural India
%
All India Average
1970
21
1980
41
1990
53
1994
64
Production
Growth
1039
1040
November 2000
any trade-offs or complementarities that may
arise between the achievement of these two
goals.
Conceptual Framework and Model
Much of the literature seeking to explain
the decline in rural poverty has focused
on agricultural growth and food price policies (Ravallion and Datt; Sen). Yet little
attention has previously been paid to the
role of government spending in alleviating poverty. Government expenditure has
not only contributed to agricultural growth
and hence indirectly to poverty alleviation,
but it has directly created rural non-farm
jobs and increased wages. The real significance of government development expenditure lies in the fact that it imparts a greater
amount of “trickle-down” benefits for the
poor in the growth process than agricultural
growth alone. Unlike agricultural growth,
which often reduces poverty only by increasing mean consumption, government expenditure reduces poverty by increasing both
mean income and improving the distribution
of income (Sen).
Another significant feature of the literature
on rural poverty in India is that most of the
previous studies have used a single-equation
approach (Ahluwalia; Saith; Gaiha; Ravallion
and Datt; Datt and Ravallion). There are
at least two disadvantages to this approach.
First, many poverty determinants such as
income, production or productivity growth,
prices, wages, and nonfarm employment are
generated from the same economic process as
rural poverty. In other words, these variables
are also endogenous variables; ignoring this
characteristic leads to biased estimates of the
poverty effects (van de Walle; Bell and Rich).
Second, certain economic variables affect
poverty through multiple channels. For example, improved rural infrastructure will not
only reduce rural poverty through improved
agricultural productivity, it will also affect
rural poverty through improved wages and
nonfarm employment. It is difficult to capture
these different effects with a single-equation
approach.
Building on previous studies of the determinants of rural poverty in India, this study
develops a simultaneous equations model
to estimate the various direct and indirect
Amer. J. Agr. Econ.
effects of government expenditures on productivity and poverty.1 The formal structure
of the system is given in equations (1) to (19).
All variables are measured in annual growth
rates (or differences in logarithm of all variables) and are defined in table 2.2
Equation (1) models the determinants of
rural poverty reduction (P).3 They include
growth in total factor productivity in agricultural production (TFP), changes in rural
wages (WAGE) and nonagricultural employment (NAEMPLY), changes in the terms
of trade (TT), changes in the percentage
of landless households in total households
(LANDN), one year lag of growth in rural
population (POP−1 ), and one year lag of
GDP growth (GDP−1 ).4 Rather than agricultural income ‘TFP’ is used in order to capture
the impact on rural poverty of technologydriven shifts in the production function,
rather than simply increased input use. Some
economists, such as Datt and Ravallion, have
used output per hectare (land productivity)
as a proxy for agricultural performance or
to represent changes in agricultural technology. But changes in land productivity do not
necessarily imply technical change because
farmers can simply use more inputs on a per
hectare basis to increase land productivity.
(1)
P = f(TFP, WAGE, NAEMPLY, TT,
LANDN, POP−1 GDP−1 )
1
Total government expenditure in India is divided into nondevelopment and development spending, and the latter is further sub-divided into spending on social and economic services.
Social services include health, labor, and other community services, while economic services include sectors like agriculture,
industry, trade, and transportation. Because we use the statelevel data in our analysis, the central government expenditure
is excluded from the study. For agriculture and allied activities,
the central government’s expenditure accounted for 32% of total
government expenditure in 1993, but most of this spending was
on fertilizer subsidies and transfers to the states for welfare programs that do not affect the capital stocks considered in this
paper. Therefore, ignoring central government spending should
not seriously bias our analysis.
2
All variables in our analysis were first transformed into
geometric annual growth rates in logarithm form, i.e., dx =
ln(xt /xt−n )/n, where xt and xt−n represent the observations on
x at time t and t − n, respectively, and n is the number of the
years between two periods when data are available. If n = 1,
then dx is simply a first difference in logarithms. This transformation avoids the problem of different time intervals between
observations. It also alleviates potential multicollinearity problems among many dependent variables on the right hand side of
the equations.
3
All variables without subscripts indicate observations in year
t at the state level. For presentation purposes, we omit the subscript. The variables with subscript “−1 · · · − j” indicate observations in year t − 1 · · · t − j.
4
Lagged instead of current population and GDP growth are
used because we can avoid the endogeniety problem of these
variables in the poverty equation. However, when the current
growth is used, the results differ very little.
Fan, Hazell, and Thorat
Table 2.
Government Spending, Growth and Poverty in Rural India
1041
Definition of Exogenous and Endogenous Variables
Exogenous Variables
POP−1
WAPI
GDP−1
ATT
TFPn
RAIN
One year lag of rural population growth.
World agricultural price index (average export price for rice, wheat and corn).
One year lag of gross domestic product.
Lagged five-years moving average of the terms of trade variable.
Total factor productivity growth at the national level.
Annual rainfall.
Endogenous Variables
IRE
RDE
ROADE
EDE
PWRE
GCSSL
GCSHEL
GERDEV
P
LITE
ROADS
IR
PUIR
PRIR
PVELE
WAGE
NAEMPLY
TFP
LANDN
TT
Government expenditure on irrigation, both from revenue and capital accounts.
Government spending (both revenue and capital) on agricultural R&D.
Government investment and spending in rural roads.
Government spending on rural education.
Government revenue and capital spending on rural power.
Government capital stock accumulated in soil and water conservation investment.
It is the weighted average of the past government expenditure on soil and
water conservation, i.e., GCSSLt = m wm SOILEt−m , where SOILEt−m is government expenditure on soil and water conservation at time t−m. The weights
are 0.4, 0.3, 0.2, and 0.1, respectively, with three years lag.
Government spending on medical and public health and family welfare measured
in stock terms using three years lag similar to expenditures on soil and water
conservation.
Government expenditure on rural and community development measured in
stock terms using three years lag similar to expenditures on soil and water
conservation.
Rural population falling below poverty line.
Literacy rate of rural population.
Road density in rural areas.
Percentage of total cropped area that is irrigated (sum of both public and private
irrigation).
Percentage of total cropped area under public irrigation (canal irrigation).
Percentage of total cropped area under private irrigation (wells, tube wells, and
tanks).
Percentage of rural villages that are electrified.
Wage rate of rural labor.
Percentage of nonagricultural employment in total rural employment.
Total factor productivity growth (Tornqvist-Theil index).
Percentage of rural households that are landless.
Terms of trade, measured as agricultural prices divided by a relevant non agricultural GNP deflator.
Wages and nonagricultural employment
are also important sources of income for
rural residents in India. Income from wages
can derive from both agricultural and
nonagricultural sources. The terms-of-trade
variable measures the impact of changes in
agricultural prices relative to nonagricultural
prices on rural poverty.5 It is hypothesized
5
Instead of using the inflation rate in rural areas (as did Saith
1981; Ahluwalia 1985; Bell and Rich 1994; Datt and Ravallion
1997), we use the terms of trade (agricultural prices relative to
nonagricultural prices). The reason is that the rise of agricultural
prices may have even greater impact than the general price index
on rural poor since they are usually net buyers of agricultural
products.
that in the short run, the poor may suffer
from higher agricultural prices because they
are usually net buyers of foodgrains. The percentage of landless households is included in
the equation to measure the potential impact
of access to land on rural poverty. Population
growth also affects rural poverty since higher
growth in population may increase rural
poverty if there is insufficient growth in rural
employment. This is particularly important
for a country like India where resources are
limited and the population base is large. The
GDP growth variable is included to control
for the remaining income effects (in addition
1042
November 2000
Amer. J. Agr. Econ.
to growth in TFP, wages, and nonagricultural
employment) on rural poverty reduction.
(2) TFP = f RDE, RDE−1 RDE−i IR, ROADS, PVELE, LITE
GCSHEL, GERDEV
GCSSL, GDP−1 RAIN
Equation (2) models the determination of
TFP growth in agriculture. The TFP growth
index is the ratio of an aggregated output
index to an aggregated input index.6 The following variables are included in the equation:
current and lagged government spending in
agricultural research and extension (RDE,
RDE−1 RDE−i ); percentage of irrigated
cropped area in total cropped area (IR); literacy rate of the rural population (LITE);
road density (ROADS); percentage of village
electrified (PVELE), capital stocks of government investment in health (GCSHEL),
rural development (GERDEV), and soil
and water conservation (GCSSL); lagged
GDP; and annual rainfall (RAIN). The first
seven sets of variables should capture the
productivity-enhancing effects of technologies, infrastructure, education, and other various government spending in rural areas. The
lagged GDP variable should control for the
effects of overall economic growth on TFP
growth in agriculture. The rainfall variable
should capture the weather effects. In the initial estimation, an effort was made to separate out the differential impacts of public and
private irrigation in the equation, but these
two variables are highly correlated. We therefore added these two variables together and
used total irrigated area as a percent of the
cropped area in equation (2).
(3) WAGE = f TFP, ROADS, PVELE
LITE, GCSHEL
GERDEV, GCSSL
GDF−1
Equation (3) is a wage determination
function. Changes in rural wages are modeled as a function of growth in agricultural
6
Another advantage of using TFP growth instead of production growth is that the TFP function has significantly fewer
independent variables than the production function. The production function includes input variables like labor, land, fertilizer, machinery, and draft animals as independent variables in
addition to those variables included in the TFP function. Fewer
independent variables in the TFP function help reduce multicollinearity problems, and increase the reliability of the estimated
results.
TFP; improvement in infrastructure (roads
and electricity) and education; and government spending on health, rural development,
and soil and water conservation; and lagged
GDP. The impact of improved infrastructure on wages is often ignored in specifying wage determination equations. Ignoring
this effect is likely to lead to underestimation of the impact of government spending
on poverty, since wage increases induced by
improved rural infrastructure can be potentially large, benefitting workers in agricultural and nonagricultural activities. The GDP
growth variable is included to control for
other factors that are closely linked to overall economic growth.
(4)
NAEMPLY = f TFP, ROADS
PVELE, LITE
GCSHEL, GERDEV
GCSSL, GDP−1
Equation (4) determines nonagricultural
employment. It is modeled as a function
of growth in agricultural TFP; development in rural infrastructure and education;
government expenditures on health, rural
development programs, and soil and water
conservation; and growth in GDP. The first
three variables measure how growth in agricultural productivity, and improvement in
infrastructure and education help to generate nonfarm employment opportunities.
Improved roads and electricity access will
help farmers to set up small non-farm businesses and to market their products out
of their villages and towns. Improved roads
and education also help farmers to find
jobs in towns. Improved health status of
farmers may also help to improve their
nonagricultural employment opportunities.
Government programs in rural development,
such as the Integrated Rural Development
Programs and rural employment schemes,
are designed by the government to alleviate
rural poverty by generating nonagricultural
and wage employment opportunities for rural
laborers. Government spending on soil and
water conservation is also often used by the
government to generate wage employment
for farmers, particularly in drought years. As
with the wages equation, we include GDP
Fan, Hazell, and Thorat
Government Spending, Growth and Poverty in Rural India
growth to capture its effect on growth in
nonagricultural employment.7
(5)
PUIR = f(IRE IRE−1 IRE−j PVELE)
Equation (5) models the relationship
between government investment and the percentage of cropped areas under public irrigation. Since most canal irrigation is supported
by the government, the cropped area under
canal irrigation is used as a proxy for public
irrigation. The explanatory variables are current and past government spending on irrigation (IRE, IRE−1 IRE−j ), and access to
electricity for pumping (percentage of electrified villages).8
(6)
PRIR = f(PUIR PVELE)
Equation (6) models the impact of public
irrigation on private irrigation. It is argued
that canal irrigation supported by the government is often a pre-condition for private irrigation to have a reasonable economic return.
The private irrigation is defined as the percentage of cropped areas under wells and
tube wells, which are mostly farmers’ private
initiatives.
(7) ROADS = f(ROADE ROADE−1 ROADE−k )
(8)
LITE = f(EDE EDE−1 EDE−m )
(9)
PVELE = f(PWRE PWRE−1 PWRE−n )
Equations (7), (8), and (9) model the
relationships between road density and
7
Again, lagged instead of current GDP growth is used for both
wages and nonfarm employment equations. The difference in
results is small whether current or lagged GDP growth is used,
however.
8
To test whether there is any difference in the impact of current and capital account expenditure, we included both a capital
stock variable (using three years lag) and a current expenditure
variable for irrigation in equation (8). The results revealed that
capital expenditure has a significant and positive effect on the
percentage of irrigation, but the current expenditure has a small,
negative, but statistically insignificant impact. This seems to indicate that government may overspend in the current account and
underspend in the capital account. Similar tests could not be
done for government expenditure on roads, education, agricultural R&D, and rural development, because these government
expenditures are mainly from current account.
1043
current and past government investment
in rural roads (ROADE, ROADE−1 ROADE−k ), between the rural literacy rate
and current and past government investment
in education (EDE, EDE−1 EDE−m ),
and between the percentage of villages
electrified and government investment in
power (PWRE, PWRE−1 PWRE−n ),
respectively.
(10) LANDN = f(TFP POP−1 NAEMPLY)
Equation (10) models the effect of productivity growth on access to land. It has often
been argued that improved productivity as
a result of technology adoption and infrastructure improvements has worsened equity
problems in rural areas in India. We expect
increased landlessness to be highly correlated with any worsening of rural poverty,
hence endogenizing access to land in the
model should capture the negative impacts
on the poor arising from productivity growth
due to technological change and infrastructure improvement. We also include non-farm
employment and rural population growth in
the equation to capture their effects on access
to land.
(11) TT = f(TFP TFPn WAPI)
Equation (11) determines the terms of
trade. Growth in TFP at the state and
national levels (TFPn) increases the supply of agricultural products, and therefore
reduces agricultural prices. The inclusion of
national TFP growth will help to reduce
any upward bias in the estimation of the
poverty alleviation effects of government
spending within each state, since TFP growth
in other states will also contribute to lower
food prices through the national market. A
weighted average of the world price index of
rice, wheat, and corn (WAPI) is also included
in the equation to capture the impact of
international markets on domestic agricultural prices.
Many have argued that government expenditures may also be endogenous, and without controlling for this, the estimates of the
model’s parameters will be biased. Two features of our model specification may help to
reduce this bias. First, annual growth rates
(instead of levels) of all variables were used.
This is similar to taking first differences (in
log form). If the factors driving government
1044
November 2000
investment are fixed over time, then the fixed
effect will be eliminated by this procedure.
This would control, for example, the influence
of agroclimatic factors on public investment
decisions. More importantly, lagged instead
of current government expenditures are used
in the government infrastructure, technology,
and education equations to capture the long
lead times involved in translating actual government expenditures into productive capital
stocks. The simultaneity between current productivity growth (or poverty reduction) and
past government expenditures is likely small
or even nonexistent. Nevertheless, just in case
any remaining endogeneities exist, we endogenize government expenditures on various
items as functions of lagged state GDP and
terms-of-trade (ATT) in equations (12)–(19).
It is expected that all coefficients are positive. Note that we do not need to control for
agroclimatic variables because these drop out
when the variables are expressed in growth
form. Double-log functional forms are used
for all equations in the system.
(12)
RDE = f(GDP−1 ATT)
(13)
ROADE = f(GDP−1 ATT)
(14)
IRE = f(GDP−1 ATT)
(15)
EDE = f(GDP−1 ATT)
(16)
GCSSL = f(GDP−1 ATT)
(17)
PWRE = f(GDP−1 ATT)
(18) GERDEV = f(GDP−1 ATT)
(19)
GCSHEL = f(GDP−1 ATT)
Data, Model Estimation, and Results
Data Sources and Measurement
We have panel data for most variables for
fourteen states from 1953 to 1993.9 However, the system was estimated only for the
period of 1970 to 1993 due to long lag effects
of expenditures on productivity growth and
poverty reduction. The years 1971, 1974–76,
1978–82, 1984–85, and 1991 were also deleted
because of missing values. A total of 154
observations was used in the final estimation.
9
The states included are Andhra Pradesh, Bihar, Gujarat,
Haryana, Karnataka, Kerala, Madhya Pradesh, Maharashtra,
Orissa, Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh, and West
Bengal.
Amer. J. Agr. Econ.
The head-count ratio, which measures
poverty as a percentage of the rural population falling below the poverty line, is used
in the analysis. Other measures, such as the
poverty-gap index, the squared poverty-gap
index, and the Sen index, are also used by
many scholars to supplement the head-count
ratio. But the head-count ratio is the most
important indicator used by the Indian government in setting its goals to alleviate rural
poverty. The head-count ratio data used in
this analysis were constructed by Gaurav
Datt and published by the World Bank
(World Bank 1997). Datt used the poverty
line originally defined by the Planning Commission and more recently endorsed by the
Planning Commission, which is based on a
nutritional norm of 2,400 calories per person
per day. It is defined as the level of average per capita total expenditure at which this
norm is typically attained and is equal to
a per capita monthly expenditure of Rs 49
at all-India rural prices for October 1973–
June 1974.
The TFP index is defined as aggregate output index minus aggregated input index and
calculated by the authors.10 The road density
variable is defined as length of road per unit
of geographic area. The education variable
is the literacy rate defined as the percentage
of the total population that is literate. Public
irrigation is defined as the percentage of the
cropped area under canal irrigation, and private irrigation is defined as the percentage
of the cropped area under well and tubewell irrigation. The electrification variable is
the percentage of villages that have access
to electricity. The rural wage used is the
10
A Tornqvist-Theil index is used to aggregate both inputs and
outputs. Specifically,
∗
∗
ln TFPt =
i 1/2 (Si t + Si t−1 ) ln(Yi t /Yit−1 )
− i 1/2∗ (Wi t + Wi t−1 )∗ ln(Xi t /Xit−1 )
where ln TFPt is the log of the total factor productivity index;
Wi t and Wi t−1 are cost shares of input i in total cost at time
t and t − 1, respectively; and Xi t and Xi t−1 are quantities of
input i at time t and t − 1, respectively. Five inputs (labor, land,
fertilizer, tractors and buffalo) are included. Labor input is measured as the total number of male and female workers employed
in agriculture at the end of each year; land is measured as gross
cropped area; fertilizer input is measured as the total amount of
nitrogen, phosphate, and potassium used; tractor input is measured as the number of four-wheel tractors; and bullock input
is measured as the number of adult bullocks. The wage rate for
agricultural labor is used as the price of labor; rental rates of
tractors and bullocks are used for their respective prices; and the
fertilizer price is calculated as a weighted average of the prices of
nitrogen, phosphate, and potassium. The land price is measured
as the residual of total revenue net of measured costs for labor,
fertilizer, tractors, and bullocks.
Fan, Hazell, and Thorat
Government Spending, Growth and Poverty in Rural India
male labor rate in real terms deflated by the
consumer price index for rural labor. These
variables were aggregated from district level
data, which were obtained from the Planning
Commission through the National Center for
Agricultural Policy and Economics Research,
New Delhi.
Nonagricultural employment is measured
as the percentage of nonagricultural
employment in total rural employment.11
Data on nonagricultural employment are
only reported by the NSS for every five years
beginning in 1973. The data for other years
were estimated by geometric interpolation.12
The terms of trade variable is measured as
the change in agricultural prices relative to
nonagricultural prices. The landless variable
is measured as the percentage of rural households classified as landless. Since the data are
only available every ten years beginning in
1953, the data for intermediate years were
estimated by geometric interpolation.
Government expenditure data by state
were obtained from Finances of State Governments, various issues, published by the
Reserve Bank of India. All the expenditures are deflated to 1960/61 prices using
a national gross domestic product deflator.13
They include expenditures from both the
current account (for maintenance) and the
capital (investment) account.
Agricultural R&D expenditure includes
government expenditure on agricultural
research and extension, while government
expenditure on irrigation includes spending
on irrigation and flood control. Government
expenditure on roads, education, and power,
and health in rural areas are calculated using
the percentage of rural population in total
population and total government expenditures on these items.
GDP at the state level is measured in
1960/61 prices and is reported in the official
11
Employment is defined on the basis of the usual status
(more than 50% of time) of workers in a particular employment
category.
12
Interpolation between available data years is reasonable
when the variable fluctuates little around its trend. This is more
likely to be true for population growth or literacy than for
employment. In order to check the reliability of the method for
employment, we re-estimated the non-farm employment equation using only the years in which non-farm employment data
are available (1972, 1977, 1983, 1987, and 1993). The coefficients
of the new estimates differed only slightly, while the t values
improved and the R2 became slightly smaller.
13
Ideally, government expenditures should be deflated using the
state GDP deflators. But the state GDP deflators are not available to us. We do have a consumer price index for agricultural
labor (CPIAL) by state over time. To conduct a sensitivity test,
we used CPIAL to deflate the government expenditures. The
final estimated results changed very little.
1045
state statistical abstracts. Table 1 summarizes
the changes over the sample period in some
of the key variables.
Tests of Investment Lags
Government investments in R&D, roads,
education, power, health, and irrigation can
have long lead times in affecting agricultural production, as well as long-term effects
once they kick in. One of the thornier
problems to resolve when including government investment variables in a production or productivity function concerns the
choice of an appropriate lag structure. Most
past studies use stock variables which are
usually weighted averages of current and
past government expenditures on individual
investments like R&D. But the choice of
weights and the length of the lags to use
remains arbitrary. We used a free-form lag
structure in our estimation procedure. Current and past values of government expenditures on different investments, such as
R&D, irrigation, roads, power, and education,
were included in the equations for productivity, technology, infrastructure, and education.
Then statistical tools were used to test and
determine the appropriate length of lag for
each investment expenditure.
Various procedures have been suggested
for determining the appropriate length of lag.
The adjusted R2 and Akaike’s Information
Criteria (AIC) are used by many economists
(Greene). We used the adjusted R2 criterion, and chose the lag lengths that maximized the R2 for each investment equation.14
This led to lags of thirteen years for R&D,
eight years for irrigation, eleven years for
education, seven years for power, and eight
years for roads. These lags are quite short
compared to much longer lags obtained for
the United States (Pardey and Craig; Alston,
Craig, and Pardey).
Another problem that arises in estimating
lag structures is that the independent variables (e.g., RDE, RDE−1 , RDE−2 and
RDE−i in the TFP function) are often highly
correlated, making the estimated coefficients
statistically insignificant. Many ways of tackling this problem have been proposed. The
most popular approach is to use what are
14
Another advantage of the single equation test is that the estimation is not constrained by missing values of the variables in
other equations in the system (for example, poverty data), leading to more reliable estimates because of the larger number of
observations.
1046
November 2000
called polynomial distributed lags, or PDLs.
In a polynomial distributed lag, the coefficients are all required to lie on a polynomial of some degree d. In this study, a PDLs
of degree 2 was used. In this case, it is
only necessary to estimate three instead of
i + 1 parameters for the lag distribution.
For more detailed information on this subject, see Davidson and MacKinnon. Once the
lengths of lags are determined, the simultaneous equation system with the PDLs and
appropriate lag length for each investment
can be estimated.15
Instead of using current and past expenditures, stock variables are used to measure the
impact of government spending on health,
rural development, and soil and water conservation. This is because these expenditures
usually have relatively quick impacts on rural
poverty. A three-year lag structure is used
with the weights of 0.4, 0.3, 0.2, and 0.1 for the
current year, t −1, t −2, and t −3, respectively.
Results
We used the full information maximum likelihood (FIML) technique to estimate the equations system. The results are presented in
table 3.
The estimated poverty equation (equation (1)) supports the findings of many
previous studies. Improvements in agricultural productivity, higher agricultural wages,
and increased non-agricultural employment
opportunities have all contributed significantly to reducing poverty, whereas improvements in the terms of trade for agriculture
significantly increase rural poverty. The incidence of landlessness and rural population
growth are positively but insignificantly correlated with rural poverty. The GDP growth
variable has a statistically insignificant impact
on rural poverty reduction. This may be
due to the fact that the impact of income
growth on rural poverty has already been
captured by other variables in the equation,
such as growth in TFP, wages, and non-farm
employment.
The estimated TFP equation (equation (2))
shows that agricultural research and extension, improved roads, irrigation, and education have all contributed significantly to
15
The sums of the coefficients from PDLs and free forms lag
structure are not significantly different for all types of expenditure except R&D. The summed coefficient of R&D expenditure
from PDLs is substantially larger than that from free form lag
structure (0.255 vs 0.090). Therefore, the estimated productivity
and poverty effects from free forms lag structure are also substantially lower than those from PDLs distribution.
Amer. J. Agr. Econ.
growth in total factor productivity. The coefficient for agricultural research and extension is the sum of the past thirteen years
coefficients from the PDLs distribution. The
significance test is the joint t test of three
parameters of the PDLs.16 On the other
hand, rural electrification, government spending on health, rural development, and soil and
water conservation are not statistically significant. The impact of state GDP growth on
growth in agricultural TFP is small; in fact,
the coefficient is statistically insignificant and
negative.
The estimated coefficients for equation (3)
show that growth in TFP and increases in
education and roads have significantly contributed to increases in rural wages, while
rural electrification, government spending on
rural development, health, and soil and water
conservation have no significant impact on
rural wages. The estimates for equation (4)
show that nonagricultural employment
increases significantly with roads and education, and government investments in rural
development. Surprisingly, nonagricultural
employment does not increase significantly
with agricultural TFP. However, the coefficient for GDP is positive and statistically
significant in the nonfarm employment equation, and this variable may be capturing the
TFP effect.
The estimates for equations (5) and (6)
confirm that government investments in
irrigation have significantly increased the
percentage of the cropped area under irrigation, and that public irrigation contributes
significantly to additional private irrigation.
The estimated results for equations (7), (8),
and (9) show that government investments
in roads, education, and power have all contributed to the stock of these variables.
The estimates for equation (10) show
that growth in TFP does not lead significantly to any increase in the incidence of
landlessness among rural households, the
so-called “immiserizing growth” phenomena. The estimated terms of trade equation
(equation (11)) confirms that increases in
TFP do exert a significant downward pressure
on agricultural prices, worsening the terms
of trade for agriculture. It also shows that
domestic agricultural prices are highly correlated with world agricultural prices.
16
The joint t test values are calculated following Greene 1993
(p. 187).
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
(16)
(17)
(18)
(19)
P = −0034 −
0.171 TFP
(−132)
(−258)∗
TFP = −0026 + 0.255 TRDE
(−078)
(182)∗
+ 0.0015 GCSSL
(0.37)
WAGE = −0035 +
0.129 TFP
(−139)
(186)∗
+ 0.273 GDP−1
(1.32)
NAEMPLY = −0029 −
0.058 TFP
∗
(−067)
(−272)
+ 0.209 GDP−1
(260)∗
PUIR = −0021 +
0.087 TIRE
(−066)
(449)∗
PRIR = 0.017 +
0.918 PUIR
(1861)∗
(223)∗
ROADS = 0.088 + 0.232 TROADE
(283)∗
(459)∗
LITE = 0.087 + 0.067 TEDE
(652)∗
(459)∗
PVELE = 0.107 + 0.072 TPWRE
(256)∗
(634)∗
LANDN = −0011 +
0.026 TFP
(−089)
(0.72)
TT = 0.025 −
0.175 TFP
∗
(−303)∗
(222)
RDE = 0.107 + 0.363 GDP−1
(0.82)
(220)∗
ROADE = 0.224 + 0.482 GDP−1
∗
(0.31)
(545)
IRE = 0.478 − 0.431 GDP−1
∗
(−053)
(520)
EDE = 0.123 + 0.336 GDP−1
∗
(179)∗
(577)
GCSSL = −0140 + 0.773 GDP−1
(2.31)
(−352)∗
PWRE = 0.133 + 1.490 GDP−1
(1.02)
(224)∗
GERDEV = 0.113 + 1.476 GDP−1
(356)∗
(249)∗
GCSHEL = 0.177 − 0.123 GDP−1
(−0224)
(577)∗
− 0.185 WAGE
(−224)∗
+
0.215 IR
(183)∗
− 0.141 GDP−1
(−097)
+ 0.231 ROADS
(228)∗
+
+
+
+
+ 0.190 ROADS −
(245)∗
0.263 TT
(2.52)∗
0.242 ROADS
(243)∗
0.272 RAIN
(547)∗
0.062 PVELE
(0.57)
− 0.594 NAEMPLY + 0.024 LANDN + 0.320 POP−1 + 0.072 GDP−1 R2 = 0113
(−312)∗
(0.43)
(0.31)
(0.82)
+ 0.062 PVELE +
0.708 LITE
+ 0.012 GCSHEL + 0.022 GERDEV
∗
(0.60)
(195)
(0.39)
(0.63)
R2 = 0301
0.045 PVELE
(−094)
+
+
0.939 LITE
(201)∗
+ 0.026 GCSHEL − 0.024 GERDEV +
(0.83)
(−088)
0.013 GCSSL
(0.74)
0.710 LITE
(327)∗
+ 0.011 GCSHEL + 0.030 GERDEV −
(0.24)
(223)∗
0.003 GCSSL
(−037)
R2 = 0093
R2 = 0311
R2 = 0087
+ 0.067 PVELE
(1.05)
+ 0.012 PVELE
(0.87)
R2 = 0697
R2 = 0147
R2 = 0277
R2 = 0028
+ 0.511 POP−1
(182)∗
− 0.792 TFPn
(−554)∗
+ 0.550 ATT
(239)∗
+ 0.534 ATT
(243)∗
− 0.254 ATT
(−061)∗
− 0.075 ATT
(−079)∗
+ 0.594 ATT
(232)∗
+
1.11 ATT
(178)∗
+ 0.677 ATT
(311)∗
+ 0.173 ATT
(0.85)
− 0.142 NAEMPLY
(−146)
+
0.271 WAPI
(803)∗
R2 = 0059
R2 = 0363
R2 = 0028
R2 = 0018
R2 = 0020
R2 = 0056
R2 = 0151
R2 = 0017
R2 = 0292
R2 = 0058
1047
Note: Coefficients for expenditures on R&D (TRDE), irrigation (TIRE), roads (TROADE), education (TEDE), and power (TPWRE) are sums of coefficients of current and lagged expenditures.
* Indicates significance at the 5% level or better. The t tests for TRDE, TIRE, TROADE, TEDE, and TPWRE were calculated for the linear combination of all the relevant coefficients of the lagged variables (Greene 1993, p. 187).
Government Spending, Growth and Poverty in Rural India
(8)
Estimated Model
Fan, Hazell, and Thorat
Table 3.
1048
November 2000
The results for the equations determining public expenditure on different items
(equations (12)–(19)) show that many government expenditures are positively and significantly correlated with GDP growth and
improvements in the terms-of-trade. But the
low R2 s obtained indicate that possible endogeneities with these variables do not significantly affect the estimated coefficients in
other equations. This may be explained by the
fact that using the first differences for all variables may have largely eliminated any endogeneity problems with the public expenditure
variables.
The estimated model shows that improvements in agricultural productivity not only
reduce rural poverty directly by increasing income (equation (1)), but they also
reduce poverty indirectly by improving wages
(equation (3)) and lowering agricultural
prices (equation (11)). On the other hand,
improvements in agricultural productivity
contribute to worsening poverty by increasing landlessness (equation (10)), though this
effect is relatively small and insignificant.
Rural Poverty Elasticities and
Marginal Impact
By totally differentiating equations (1)
to (11), we can derive the marginal impacts
and elasticities of different types of government expenditures on productivity growth
and rural poverty reduction after allowing
for all relevant direct and indirect impacts.
These can be viewed as short- to mediumterm impacts because we ignore the impact of
increases in income on future levels of public
expenditure.
The impacts of different types of
government spending on rural poverty and
agricultural productivity are shown in table 4.
Two impact measures are presented. The
first measure is the elasticity of each item
of government spending, and this gives the
percentage change in poverty or productivity
corresponding to a 1% change in government expenditure on that item. Since all
expenditures are measured in rupees, then
these elasticities provide a measure of the
relative growth and poverty reducing benefits that arise from additional expenditures
on different items, where the increases are
proportional to existing levels of expenditure.
The second measure is the marginal return
(measured in poverty and productivity units)
Amer. J. Agr. Econ.
for an additional Rs 100 billion of government expenditure. This measure is directly
useful for comparing the relative benefits
of equal incremental increases in expenditures on different items, and it provides crucial information for policymakers
in setting future priorities for government
expenditure in order to further increase
productivity and reduce rural poverty. The
marginal returns were calculated by multiplying the elasticities by the ratio of the poverty
or productivity variable to the relevant government expenditure item in 1993. Table 4
also shows the number of poor people who
would be raised above the poverty line for
each Rs 1 million of additional investment in
an expenditure item.
An important result in table 4 is that
there are sizable differences in the productivity gains and poverty reductions obtained
for incremental increases in each expenditure
item. Government expenditures on roads and
R&D have far larger impacts on productivity growth and rural poverty reduction
than any other investments, and both are
attractive win-win investments. Roads expenditure has a bigger impact on poverty than
R&D (45% larger), but R&D has the biggest
growth impact (154% larger). If the government were to increase its investment in
roads by Rs 100 billion (at 1993 constant
prices), the incidence of rural poverty would
be reduced by 0.65%. Moreover, for each one
million rupee increase in investment in roads,
124 poor people would be lifted above the
poverty line. This compares with eighty five
people raised above the poverty line if an
additional million rupees is invested in R&D.
Investment in roads reduces rural poverty
through productivity growth, but also
through increased nonagricultural employment opportunities and higher wages.
Because the marginal impact is calculated as
a total derivative of the equations system,
it can be decomposed into its various direct
and indirect components (Fan, Hazell, and
Thorat). The productivity effect accounts for
31% of the total impact on poverty, nonagricultural employment for 49%, and the
remaining 20% is accounted for by increases
in rural wages. Of the total productivity
effect on poverty, 75% arises from the direct
impact of roads in increasing incomes, and
the remaining 25% arises from lower agricultural prices (17%) and increased wages
(8%). Increases in the incidence of landlessness arising from productivity growth has no
Fan, Hazell, and Thorat
Table 4.
Poverty and Productivity Effects of Government Expenditures
Elasticities
Expenditure
variable
−0060
−0009
−0050
−0053
−0003
−0001
−0019
−0001
t value
−206∗
−196∗
−255∗
−364∗
−164
−152
−368∗
−113
TFP
0255
0036
0057
0047
0004
00015
0022
0012
t value
182∗
223∗
269∗
263∗
064
037
063
039
Poverty (% Point)
TFP (% Point)
No. of Poor Reduced
(per 100 billion rupees at 1993 prices)
Rank
Rank
−045
2
601
1
−005
7
061
4
−065
1
237
2
−022
3
062
3
−0003
8
012
8
−012
5
043
6
−013
4
049
5
−009
6
038
7
(per million rupees)
Rank
845
2
97
7
1238
1
410
3
38
8
226
5
255
4
178
6
Note: Numbers in parentheses are ranks. “*” indicated significance at the 10% level. The t values are estimated using a procedure proposed by Greene (1993, p. 221). A liner Taylor series approximation to
the nonlinear restriction is used, using the large-sample properties of the estimates.
Government Spending, Growth and Poverty in Rural India
R&D
Irrigation
Road
Education
Power
Soil and Water
Rural Development
Health
Poverty
Marginal impact
1049
1050
November 2000
significant impact on rural poverty. One reason that R&D investments have a smaller
impact on poverty is that they work entirely
through improved productivity; there is no
direct expenditure benefit for the poor.
Government spending on education ranks
as a distant third in its impact on rural
poverty and agricultural productivity. An
additional one million rupees invested in education raises forty one people above the
poverty line. Most of this effect is from
more nonfarm employment opportunities and
increased wages. Government expenditures
on rural development and soil and water conservation have the fourth and fifth largest
impacts on poverty reduction, respectively,
and an extra one million rupees of either
investment would raise a little over twenty
people above the poverty line. In addition,
these types of expenditure have no statistically significant productivity impacts, and
hence do not contribute to long-term solutions to the poverty problem. In fact, in
many regions, these types of investments were
mainly used by the government to alleviate
poverty, particularly during the drought years.
Government expenditure on health ranks
sixth in terms of its impact on poverty;
each one-million-rupee increase in expenditure on health raises eighteen poor people
above the poverty line. Irrigation expenditure
ranks seventh in terms of its impact on rural
poverty, but ranks rather higher (fourth) in
terms of its impact on productivity growth.
Government expenditure on power has a
positive but negligible impact on both rural
poverty reduction and productivity growth.
This may be because the government has
already invested heavily in rural electrification and the marginal returns from additional
investments are now low. Not only is the
size of power expenditure relatively large in
the government’s budget (50% greater than
road expenditure in 1993), but current (operational) expenditure has also increased enormously since 1990.
Conclusions
Using state-level data for 1970 to 1993, a
simultaneous equations model was developed
to estimate the direct and indirect effects
of different types of government expenditure on rural poverty and productivity growth
in India. The results show that government
spending on productivity-enhancing investments, such as agricultural R&D and irrigation, rural infrastructure (including roads and
Amer. J. Agr. Econ.
electricity), and rural development targeted
directly to the rural poor, have all contributed
to reductions in rural poverty, and most have
also contributed to growth in agricultural
productivity.17 But their poverty and productivity effects differ greatly.
Government expenditures on roads and
R&D have by far the largest impacts on
poverty reduction and growth in agricultural productivity; they are attractive win-win
strategies. Government spending on education has the third largest impact on rural
poverty and productivity growth. Irrigation
investment has had only modest impacts on
growth in agricultural productivity and rural
poverty reduction, even after allowing for
trickle-down benefits.
Government spending on soil and water
conservation, and on rural and community development, including the Integrated
Rural Development Program, has successfully helped reduce rural poverty, but its
impact has been smaller than expenditures on
roads, agricultural R&D, and education. Government health investment had no impact on
productivity growth, and its effect on poverty
alleviation through wage increases was also
very small.
The results of this study have important
policy implications. To reduce rural poverty,
the Indian government should increase its
spending on rural roads and R&D. These
types of investment not only have a much
larger poverty impact per rupee spent than
any other government investment, but also
generate higher productivity growth. R&D
investments have a larger growth impact than
roads, but their poverty impact is smaller.
But perhaps even this tradeoff could be
reduced if R&D were targeted more specifically on the problems of poor farmers and
regions than it has been in the past. On
the other hand, government expenditures on
rural development are an effective way of
helping the poor in the short term, but
because they have little impact on agricultural productivity, they contribute little to
long-term solutions to the poverty problem.
17
The results we obtained from this study differ sharply from
Datt and Ravillion (1997) who used state aggregate development expenditures and found insignificant correlation with rural
poverty reduction. In another study, Sen (1997) found that while
the aggregate state expenditures have a positive and significant
impact on rural poverty, he failed to obtain similar results when
using individual items of government expenditures. This may be
due to the different specifications of the models.
Fan, Hazell, and Thorat
Government Spending, Growth and Poverty in Rural India
[Received February 1999;
accepted March 2000.]
References
Ahluwalia, M.S. “Rural Poverty, Agricultural
Production, and Prices: A Reexamination.”
Agricultural Change and Rural Poverty.
J. Mellor and G. Desai, eds. Baltimore MD:
Johns Hopkins University Press, 1985.
Alston, J., B. Craig, and P. Pardey. “Dynamics in the
Creation and Depreciation of Knowledge, and
the Returns to Research.” EPTD Discussion
Paper No. 35. Washington DC: International
Food Policy Research Institute, 1998.
Bell, C., and R. Rich. “Rural Poverty and
Aggregate Agricultural Performance in Postindependence India.” Oxford Bulletin of Economics and Statistics 56(May 1994):111–33.
Binswanger, H., S. Khandker, and M. Rosenzweig.
“How Infrastructure and Financial Institutions Affect Agricultural Output and Investment in India.” Washington DC: World Bank
Working Paper No. 163. World Bank, 1989.
Datt, G., and M. Ravallion. “Why Have Some
Indian States Performed Better than Others at Reducing Rural Poverty?” FCND
Discussion Paper No. 26. Washington DC:
International Food Policy Research Institute,
1997.
Davidson, R., and J. MacKinnon. Estimation and
Inference in Econometrics. New York and
London: Oxford University Press, 1993.
Fan, S., P. Hazell, and S. Thorat. Linkages
between Government Spending, Growth and
Poverty in Rural India. Research Report 110,
International Food Policy Research Institute,
Washington DC, 1999.
Gaiha, R. “Poverty, Agricultural Production and
Price Fluctuations in Rural India - A Reformulation.” Cambridge J. of Econ. 13(June
1989):333–52.
1051
Ghose, A.K. “Rural Poverty and Relative Prices
in India.” Cambridge J. of Econ. 13(June
1989):307–31.
Greene, W.H. Econometric Analysis. Englewood
Cliffs NJ: Prentice-Hall, Inc., 1993.
Griffin, K., and A.K. Ghose. “Growth and Impoverishment in the Rural Areas of Asia.” World
Develop. 7(April/May 1979):361–84.
Misra, V.N., and P. Hazell. “Terms of Trade,
Rural Poverty, Technology and Investment:
The Indian Experience, 1952–53 to 1990–
91.” Econ. and Polit. Weekly 31(March
1996):A2–A13.
Pardey, P., and B. Craig. “Causal Relationship
Between Public Sector Agricultural Research
and Output.” Amer. J. Agr. Econ. 71(February
1989):9–19.
Ravallion, M., and G. Datt. “Growth and
Poverty in Rural India.” World Bank Policy Research Working Paper. Washington DC:
World Bank, 1994.
Saith, A. “Production, Prices and Poverty in
Rural India.” J. of Develop. Stud. 17(January
1981):196–214.
Sen, A. “Agricultural Growth and Rural Poverty.”
Growth Employment and Poverty Change and
Continuity in Rural India. G.K. Chadha and
Alakh N. Sharma, eds. Delhi, India: Indian
Society of Labor Economics, 1997.
Srinivasan, T.N. “Agricultural Production, Relative
Prices, Entitlements, and Poverty.” Agricultural Change and Rural Poverty. J. Mellor and
G. Desai, eds. Baltimore MD: Johns Hopkins
University Press, 1985.
van de Walle, D. “Population Growth and
Poverty: Another Look at the Indian Time
Series Data.” J. of Develop. Stud. 21(April
1985):423–39.
World Bank. India Public Expenditure Review:
Sector Report III: Agriculture. Washington
DC, 1993.
World Bank. India: Achievements, and Challenges
in Reducing Poverty. Report No. 16483–IN,
Washington DC, 1997.