|
Agricultural Outcomes
|
|
We examine the impact of dam construction on gross and net
measures of irrigated and
cultivated area. The net measures account for the relevant area
at a single point in the
year, while the gross variables account for each separate use of
the same area during a year.
|
|
Multi-cropping is measured as the ratio of gross to net
cultivated area.
Panel A of Table 5
provides OLS estimates (equations (3) and (4)), and Panel B 2SLS
estimates. Both sets of estimates suggest no significant impact
of dams on gross or net
irrigated area in the districts where they are built (columns
(1)-(4)). The 2SLS estimate
|
|
for own district irrigated area is positive but insignificant.
The absence of a clear effect
in the district where the dam is built suggests that the
submergence and degradation of
land around the reservoir has limited irrigation gains in the
vicinity of the dam. The large
standard errors potentially reflect variation in the extent of
submergence associated with
different sizes of dam.
|
|
Dams significantly
increase gross and net irrigated area in districts located downstream.19
The 2SLS estimates exceed the OLS estimates, but both are
significant and statistically
indistinguishable from each other. The point estimate suggests
that an additional dam
increases irrigated area (gross or net) in the downstream
district by roughly 1.1%.
|
|
In columns (5)-(8) we
examine gross and net cultivated area. The 2SLS estimates of the
effect of a dam on both measures is negative and significant at the 10%
level for downstream
districts, and in columns (6) and (8) for own district. This
suggests the causal effect of a dam
is to reduce cultivable area, perhaps due to the submergence
associated with construction of
reservoir and canals. In addition, some land may be lost due to
waterlogging and salination.
|
|
Consistent with the idea that these effects are greater around
the reservoir, we observe that
the own district effect is five times as large as the effect downstream. One
explanation for
the insignificant OLS estimates is that more dams were allocated
to districts where land
availability was otherwise expanding (say, due to higher returns
to agriculture).
|
|
Columns (9)-(10)
examine the extent of multi-cropping and find a positive, but insignifi-
cant, increase in multi-cropping in the downstream districts.
Finally in columns (9)-(10) we
observe that, as expected, area under water-intensive High
Yielding Varieties (HYV) crops
increases downstream.
|
|
Table 6 uses annual data for 1971-1987 to examine agricultural
production and yield for
|
|
nineteen crops, and fertilizer use. Columns (1) and (2) consider
total production. Both
the OLS and 2SLS estimates suggest that dam construction led to
an insignificant decline
in overall production in the district where they were built, and
a significant increase in
production in the downstream districts. Similarly, agricultural
yield showed an insignificant
|
|
decline in the district where dams were constructed, but a
significant rise downstream (see
columns (3) and (4)). The own district results, again, suggest
that the land around dams is
degraded by dam construction. This degradation is, in part,
compensated for by increased
productivity elsewhere in the district. The downstream districts
that do not bear any of
the environmental costs associated with dam construction enjoy
positive productivity gains.
|
|
Finally, fertilizer use increases in downstream districts, (see
column (6)).
|
|
Dam irrigation increases area devoted to water-intensive HYV
crops. In Table 8 we
|
|
examine crop-wise outcomes for six major crops, three of which
are water-intensive.20
Columns (1)-(4)
consider relatively less water-intensive crops, and columns (5)-(8) water-
intensive crops. We observe no impact on area devoted to different crops in the dam’s
own district. In downstream districts we observe a weak positive
increase in total area
|
|
devoted to water-intensive and non water-intensive crops, and a
very significant increase in
the area devoted to wheat, sugar and rice. Wheat and rice saw a
sharp increase in water-
intensive HYVs over our sample period. Sugarcane is a
water-intensive cash crop which is
very important in Western India (where most dam construction is
concentrated).
|
|
The impact of dams on
crop yield is modest, even for highly water intensive crops (Panel
B). The increase in area devoted to water intensive crops,
combined with modest yield in-
creases, leads to a significant increase in the production of
water intensive crops in the
downstream district. An additional dam increases production of
water intensive crops down-
stream by 0.6%. This is mainly attributable to a large increase
in sugar and rice production.
|
|
However, millet and wheat production also increases significantly
in the downstream district.
These results are
significant given that a major claim of dam critics is that dams cause
farmers to substitute towards water intensive crops which, while
more profitable in the short
run, accentuate water shortages in the long run. While we find
evidence that the area and
|
|
production of crops using more water, HYV crops and sugarcane,
increase in downstream
districts, we do not observe any significant substitution away
from major non water-intensive
|
|
crops.
|
|
Moreover, the area and
production of non-water intensive crops shows an insignificant
decline in the district where the dam is placed whereas
production of water-intensive crops
increases. This pattern of coefficients is sensible given that irrigated
area shows some increase
while overall cultivated area in these districts declines.
However, these results must be
|
|
interpreted with caution, since none of the individual coefficients are significant.
|
|
Our results provide a consistent picture of the impact of dams
on agricultural outcomes.
|
|
In the districts where they are built dams do not significantly
alter overall agricultural
production. In downstream districts, they enhance overall
agricultural production, and
production of some water-intensive cash crops (sugar) and
staples which have seen the advent
of HYV (wheat and rice).
|
|
|
|
Other inputs
|
|
The change in crop mix and the increase in HYV seed and
fertilizer use in downstream areas
is consistent with the predictions of a simple agricultural
production function.
The two other inputs
that the agricultural production function suggests should be af-
fected by increased dam irrigation are the use of alternative
forms of water infrastructure
|
|
(this should decline), and the use of electricity (this should
increase because pumping water
through the canals associated with dams requires electricity).21
Panel A of Table 8
examines the impact of dams on the incidence of different forms of vil-
lage infrastructure (for brevity we only report 2SLS estimate).
The results are as expected;
in downstream districts electrification increases and non
dam-related water infrastructure de-
|
|
creases (the number of canals shows an insignificant increase).
There are no other significant
implications of dam construction for public good provision in
own district or downstream,
suggesting no crowding in (or out) of other government inputs.
|
|
|
|
Cost Benefit Analysis of Dams
|
|
We use our above results to provide a cost-benefit analysis of
dams. Our analysis is tentative
since it requires several clearly contestable assumptions and is
based on somewhat noisy point
estimates. In order to obtain an upper bound on this cost benefit
analysis we choose the
assumptions which are the most favorable to dams.
|
|
We start with estimating the extent to which farmers substitute
dam irrigation for other
|
|
forms of irrigation. Our estimates suggest that a dam increased
net irrigated area by 0.7%
in its own district, and 1% downstream. Combined with the size
of irrigated area in the
average district, these point estimates imply a dam-induced net
increase in irrigated area of
6,300 hectares in 1985 (row B5 in table 9). Using the Indian
agricultural census the Planning
|
|
Commission estimated that, in 1985, 23.6 millions hectares were
irrigated by dams, or 8,758
hectares per dam (cited in Thakkar (2000)). These estimates
suggest a modest crowding-out
of other investment in irrigation (of the order of 30%: see row
B6 in Table 9).
The cost of dam
construction is generally expressed in terms of the cost of irrigating
an additional hectare by dam. We therefore base the cost-benefit
analysis on a comparison
|
|
of the value of additional production per additional hectare
irrigated (from our estimates)
with the capital and recurrent cost of an additional hectare
irrigated by a dam (Planning
Commission estimates, cited in Thakkar (2000)). Using 1985 means
and our estimates, we
calculate the increase in production due to a dam in a district
downstream to be Rs. 2.99
|
|
million annually, or Rs. 60 million in present discounted value
(assuming a 5% discount
rate, and an infinite life span for the dam). This is due to an
increase in irrigated area of
3,864 hectares. The present discounted value of the net increase
in production per irrigated
hectare is Rs. 13,686. In addition, farmers face a lower
irrigation cost (some farmers who
would have used ground water irrigation now use dam irrigation).
Since a farmer always had
|
|
the option of not using irrigation, an upper bound on reduction
in irrigation cost is the value
of increased production on the land. As discussed earlier, our
estimates suggest that dams
substituted for other forms of irrigation in 30% of dam
irrigated land. Therefore, we divide
our benefit estimate by 70% to obtain the net benefit of the dam
on agricultural profits (Rs.
|
|
19,011 in row C7).
The Planning Commission
estimated the development cost per hectare of dam irrigation
at Rs. 16,129 in 1985 (inclusive of capital cost, and an annual
maintenance cost of Rs. 300
per hectare). Adding the fertilizer cost we obtain a total cost
of Rs. 18,807 per hectare.
|
|
Without accounting for
the deadweight loss of taxation, these estimates suggest a barely
positive net present value of dam construction (1%). This turns
negative if we assume
a conventional 15% figure for the deadweight loss associated with
raising funds through
taxation. Our calculation overestimates the economic value of a
dam in so far as it does
not account for the production decline in the dam’s own district
(while imprecise and not
|
|
significantly different from zero, the point estimate is large and negative), and
additional
labor and environmental costs. It underestimates the economic
value of a dam in that it does
|
|
not account for production gains in non-agricultural sectors due
to electricity generation by
|
|
multipurpose dams.
Our estimates for the
production impact of each dam, combined with the annual increase
in the number of dams over the period (about 0.4 per year),
indicates that dam construction
was responsible for about 9% of the growth in agricultural
production between 1971 and 1987.
|
|
The World Commission on Dams (2000b), using very different methods,
attributed 10% of
the growth in India’s agricultural production since 1950 to
dams, and concluded that the
average dam’s net present value was slightly negative. Although
this estimate was made by
a supposedly independent, non-partisan international body, dam
proponents criticized these
estimates as overly conservative. The International Commission
on Large Dams (ICOLD),
|
|
an international body for dam builders, claimed in its response
that the contribution of
large dams to the growth in agricultural production was closer
to 80% (Gopalakrishnan
2000). Our estimates suggest that World Commission on Dams
(2000b) estimates are closer
to the truth.
|
|
|
|
Rural welfare
|
|
In this section we examine whether dams have created population
groups who have not
received adequate compensation for losses suffered, or whether the
productivity effects of
|
|
dam construction, combined with appropriate redistributive
policies, prevented the creation
of such groups.
We use a district
panel data-set on rural consumption, poverty and wage outcomes.
Clearly, our results would be biased if dam construction induces
either the relatively rich
|
|
or the relatively poor populations to migrate across district
boundaries. Panel B of Table 8
estimates the impact of dam construction on district census
rural population outcomes. Dam
construction has a small insignificant positive effect on overall
population and in-migrants
in both the dam’s own district and downstream district. This
suggests that dam-induced
population movements across the district boundaries are not a
serious concern, and we can
|
|
use the district panel to study the impact of dam building on
rural welfare.22 However, to
assess the extent of any possible bias due to migration we
compute bounds on the influence
of migration on the estimated impact of dams on poverty. The
results are in Table 10; Panel
A provides OLS estimates and Panel B 2SLS estimates.
|
|
In columns (1) and (2) we consider mean per-capita expenditure.
In column (1), we
|
|
find that an additional dam causes a statsitically significant
decrease of 0.3% in per-capita
|
|
expenditure in the OLS specification, and an insignificant decline
of 0.35% in the 2SLS spec-
ification. Column (2) includes dams built in upstream districts
as an additional explanatory
variable. The coefficient on the expenditure in a dam’s own
district remains negative, and
is now significant at the 10% level in the 2SLS regression (and
at the 5% level in the OLS
|
|
regression). Dams have a modest positive impact on per-capita
expenditure in downstream
districts. However, both the OLS and 2SLS estimates are
insignificant. The 2SLS estimate
is almost twice as large as, but statistically indistinguishable
from, the OLS estimate.
In columns (3) and (4)
we observe that the decline in per-capita expenditure translates
into an increase in poverty. Columns (3) and (4) consider the
head-count ratio (this is
|
|
the fraction of population with expenditure levels that place
them below the poverty line).
Dams significantly increase poverty in their own district, and
lead to a decline in poverty in
downstream districts. The downstream effect is significant at
5% in the 2SLS estimates. In
column (3), the OLS estimate for the own district effect is positive and
significant, but the
|
|
2SLS estimate, while positive, is insignificant: this parallels
our reduced form estimate.
Columns (5) and (6)
provide bounds that account for migration. We take the point
estimate of the effect of dams on in-migrants and make
alternative assumptions about their
poverty status. We then recompute the head-count ratio (we
follow the idea of “Manski
bounds” (Manski 1990)). In column (5), we compute the head-count
ratio assuming that all
|
|
in-migrants are poor. This reduces the head-count ratio more in
districts with more dams.
The OLS estimate of the own district effect changes sign and
is insignificant. However,
the confidence intervals in columns (5) and (4) overlap. The 2SLS
estimate is positive
but insignificant. In column (6), we recompute the head-count
ratio assuming that all in-
|
|
migrants are rich. Not surprisingly, both the OLS and 2SLS
estimates in column (6) exceed
the original estimates. As usual, as the Manski bounds are not
tight, and the results have to
be interpreted with caution. However, they do not suggest severe
bias with our 2SLS results.
The head-count ratio
is a relatively crude measure of the extent of poverty. The poverty
|
|
gap measures the depth of poverty – specifically, how much income
would be needed to bring
the poor to a consumption level equal to the poverty line.
Columns (7) and (8) consider
poverty gap as the dependent measure, and find similar effects: dams significantly
increase
the poverty gap in their own district, and significantly reduce
it downstream. The point
estimate for the poverty reduction associated with dam
construction upstream varies from
|
|
one-fourth (in the OLS) to one-eighth (in the 2SLS) of the
poverty increase in the dams’ own
district. In our sample there are, on average, 1.75 districts
downstream of each dam. This
|
|
implies that the poverty reduction in districts downstream from
the district where a dam is
|
|
constructed is too small to compensate for the poverty increase
in the dam’s own district.
In columns (9) and
(10) we find no significant effect of dam construction on inequality in
either own or downstream districts. Columns (11) and (12)
consider male agricultural wages.
Annual data are available for 1971-1994, but for fewer states
than in the poverty sample.23
|
|
One might expect higher land productivity (especially from the
production of cash crops)
to translate into higher agricultural wages. Wages increase in
districts located downstream
from a dam. The 2SLS and OLS estimates are similar and suggest
that 10 dams located
upstream cause an increase in agricultural wages of roughly
0.02% (the 2SLS estimates are
insignificant). The point estimate for own district is positive
but imprecise.
|
|
Finally, columns (13)
and (14) examine the claim that dams increase the incidence of
malaria and other waterborne diseases in neighboring areas. We
use data on annual malaria
|
|
incidence from 1976 to 1995, but find no evidence of increased
malaria incidence. This
suggests that the poverty increase is more likely related to the
loss of agricultural land and
displacement than to negative health effects.
|
|
Dams and Rainfall Shocks
A different channel through which dams may affect rural welfare is by improving water
security in the event of floods or droughts. If in years of bad
rainfall dams provide insurance
within their own district, then they may increase welfare even
if, on average, they reduce
consumption and increase poverty.
|
|
In Table 11 we use
annual rainfall data for Indian districts to examine the role of dams
in mediating the effect of rain shocks. Our rain shock
measure is the fractional deviation of
annual rainfall from the district’s historical average.24
The odd columns in
Table 11 document the effect of rain shocks on agricultural and wel-
|
|
fare outcomes. Negative rain shocks decrease area irrigated,
total production, and increase
poverty. In the even columns of Table 11 we examine whether dams
mitigate or accentuate
the effect of rain shocks (for brevity, we only report 2SLS estimates, but
the OLS estimates
are very similar). Having a dam upstream reduces the adverse effect of a negative rain
shock:
|
|
the coefficients on dams-rain shock interaction variable and rainshock variable
have the op-
posite sign. In contrast, dams amplify the effect of a bad rain shock in their own district;
the coefficients on dams-rain shock interaction variable and the rainfall
variable now have
|
|
the same sign. The amplification effect is potentially due to restrictions
on water use in the
|
|
dam’s catchment area.25
These findings have
significant implications for the dynamics of poverty in these districts.
Low level of migration and closed markets imply an amplification
of negative shocks (Jay-
achandran 2004). The poor in India have limited access to
insurance against risk in rural
|
|
India (Morduch (1995)), and faced with limited insurance options
the poor make inefficient
investments (Rosenzweig and Binswanger (1993), Rosenzweig and
Wolpin (1993), Morduch
(1995)), which may further increase poverty.
|
|
Institutions and Poverty
|
|
The inability or
unwillingness of those who benefit from dams to compensate groups of
losers, or of the government to force them to do so, when both
groups are clearly identifiable
ex-ante suggests poor institutions of redistribution.
To explore this possibility we build upon
recent work by Banerjee and Iyer (2005).
|
|
Banerjee and Iyer
(2005) demonstrate significant differences in the ability of the popu-
lation to organize, and obtain public goods, across Indian
districts. They argue that these
differences stem, in part, from different historical legacies. During the
colonial period, the
British instituted different land revenue collection systems
across districts. In some districts,
|
|
an intermediary (landlord) was given property rights for land
and tax collection responsibil-
ities. In other districts, farmers were individually or
collectively responsible for tax collec-
tion. ”Landlord” districts saw the emergence of a class of
landed gentry, who had conflictual
relationships with the peasants. In these districts class
relations remain tense, rendering
collective action more difficult. Districts under the landlord system
continue to have lower
|
|
public good provision, lower agricultural productivity, and more
infant mortality.26
If the politically
and economically disadvantaged are more able to demand redistribution
in non-landlord districts (and the elite feel more compelled to
compensate losers), then the
poverty impact of dams should be smaller in those districts.
Using Banerjee and Iyer’s data,
|
|
we interact dams with either a dummy for being a non-landlord
district, or the fraction
of land under non-landlord rule in a district and include this
interaction as an additional
explanatory variable (our instrument set is as before, plus
their interaction with the landlord
variable). The results are presented in Table 12, and are
striking. There is no systematic
|
|
pattern in the production and irrigation regressions, suggesting
that technology drives the
|
|
differential impact of dams on production and irrigation. However, the
impact of dams on
poverty in own district is halved in non-landlord districts (the
interaction coefficient has a
t-statistic of 1.89 in the poverty gap regression), and we
cannot reject the hypothesis that
dams do not increase poverty in non-landlord districts. There is
no significant pattern in
|
|
the downstream districts where no losers are created.
We conjecture that in
non-landlord districts the population is either more effective in
organizing to demand compensation, or more equal in sharing
among losers and winners
within the district. It is also possible that the absence of the
landed gentry gives the displaced
more political power. More generally, these findings point to the
relevance of the institutional
|
|
framework within which public policies, such as dam
construction, are executed and suggest
‘weak institutions’ or social conflict may help explain why dam
construction has particularly
strong distributional and poverty implications in developing
countries.
|
|
|
|
Robustness Checks
|
|
We conclude this section with some alternative specifications and
robustness checks
Leads and lags
Columns (1), (4), (6) and (8) in Table 13 examine whether dams affect economic outcomes
prior to their construction. This specification check is
particularly relevant for the poverty
regression since one could potentially attribute the poverty
findings to a return to normal
|
|
poverty levels following a decline while the dam was being
constructed. To examine this,
we include dams built up to 5 years in the future as an
additional variable. Our faith in
our identification assumption is bolstered by the finding that in
no case does future dam
construction affect current outcomes.
|
|
Columns (2), (5), (7)
and (9) examine whether the effect of dams persists 5 years after
dam construction. We find some evidence that the effect of dam is gradual;
dams built 5
years ago, for example, affect poverty more than those built today.
Column (3)
investigates whether dams recharge the groundwater table. If they do, then,
relative to the short run, irrigation potential in a district
should increase by more in the
|
|
long run. We estimate a level specification for 1994, where we
regress gross irrigated area in
1994 on the number of dams built in 1974 and the number of dams
in 1994 (see column (1),
Table (3) for the corresponding first stage). We cannot reject
the hypothesis that 1974 dams
do not affect 1994 irrigated area and are, therefore, unable to find evidence
of groundwater
|
|
recharging.
|
|
Functional form
Table 14 examines whether the effect of dams varies with dam size. Our
instruments are
too weak for us to estimate multiple parameters. But since the
2SLS and OLS estimates are
similar for most specifications, we feel confident about focusing
on OLS estimates.
|
|
Our regressions
include as separate regressors the number of small (below 16 meters),
medium (16 to 30 meters) and large (above 30 meters) dams. The effect of dams on poverty
is driven by large dams, while the impact of dams on
productivity is driven by medium
dams. This is in line with the case-study literature which
suggests the negative impact of
dams is the most pronounced for very large dams. It is also
possible that, for small dams,
|
|
the positive and negative effects on poverty occur within the same
district (since both the
catchment area and command area are smaller) and averaging these
effects over the district
implies no aggregate effect.
We also estimated, but
do not report, a specification which includes as separate regressors
|
|
a dummy for whether there is at least one dam in a district, and
the number of dams. The
dummy for ”at least one dam” is insignificant, and the coefficient on the number of
dams
remained unchanged (in magnitude or size) suggesting that the effect of the number of
dams
on economic outcomes is linear in the number of dams.
Alternative neighborhood measures
|
|
Table 15 examines
whether the impact of dam construction in a district extends beyond
its neighboring downstream districts. For brevity, we only
report 2SLS estimates.
In Panel A we examine
all neighboring districts. The effect of dams constructed down-
stream to a district is negative but insignificant, and there is
no effect in neighboring districts
|
|
which are neither upstream or downstream. In Panel B we examine
whether a district ben-
efits from dam construction in districts which are upstream to
its upstream neighbors and
find no effect. Clearly, some large dams may help control floods several hundred
kilometers
downstream. But for the average dam, it is unlikely that any effect extends beyond the
next
|
|
two districts, and these results suggest that it does not extend
beyond the next district.
This table suggests
that our focus on the district in which the dam was built and the
adjacent downstream district is appropriate for capturing the effects of most dams, even
if
some dams affect more than the immediate neighboring districts. It also suggests
that our
estimate of the economic impact of dams in own and upstream
districts on district outcomes
|
|
is a reasonable approximation of the overall effect of dams, save any
general equilibrium
effect which affects all district equally.
|
|
The most likely general equilibrium effect is a price effect, due to the
increase in pro-
|
|
duction. If the poor are net sellers of agricultural products
(Deaton (1989)), a decrease in
food prices is likely to accentuate poverty. In this sense, the
failure to account for general
|
0 comments: