Title: “AJAE appendix for Estimating the Impact of Welfare Programs on Agricultural Output:

Evidence from India”

Author: Girish Bahal

Date: 9 May 2019

Note: The material contained herein is supplementary to the article named in the title and published
in the American Journal of Agricultural Economics (AJAE).

1
2 Transfer Multipliers in India

Appendix A: Narrative Analysis

Below I discuss program by program construction of the narrative shock series. I express the exoge-
nous change in real per capita transfer spending (in rupees) as ∆si,t . Xi,t denotes the expenditure
incurred under the program ‘X’ in state i and year t. For ease of reference, any expenditure es-
timates that are cited while discussing the programs are the same as mentioned in the narrative
sources (in nominal prices). All expenditure estimates used in the empirical analysis, however, are
converted to 2004 prices using the state GDP deflator to control for differential price trends among
states. Ministry of Rural Development is hereafter referred to as MORD.

Integrated Rural Development Program (IRDP)

Implemented: 1978; Until: 1998

Introductory variation does not contribute to the shock series

Reason: State-wise data not available before 1985

IRDP was arguably the first major rural development program that was launched in 1978. The
program cost was equally shared between the central government and the states. The objective
of the program was to alleviate rural poverty by “providing income generating assets and access
to credit and other inputs” (Annual Report MoRD 1980, p. 3) to “all persons who live below the
poverty line” (p. 4). The program was initially introduced in 2,300 blocks in the country but
was “extended to all the development blocks in the country” (p. 4) by 2nd October 1980. The
principal motivation to introduce the program was therefore not output or growth related. Hence
the increase in transfer spending during program introduction can be regarded as an exogenous
variation. However, since state-wise data for IRDP is not available before 1985, IRDP does not
contribute to the construction of the narrative shock series.

National Rural Employment Program (NREP)

Implemented: 1980; Until: 1988

Program expenditure at the national level:

1980: |11.97 billion (2004 prices)
3 Transfer Multipliers in India

Exogenous change in state-level transfers:

∆si,1980 : N REPi,1980

To tackle the serious problem of rural unemployment and under-employment, the food for work
program was launched on the 1st of April 1977 (Annual Report MoRD 1980-81, p. 17). The
report also quotes that although the program was “successful in achieving its basic objectives ... a
number of shortcomings and drawbacks were noticed in the implementation of the food for work
program” (p. 18). With the motivation to address the shortcomings in the food for work program,
NREP “replaced the Food for Work Program in October, 1980” (Annual Report MoRD 1982-83, p.
30) and was financed by the central government up till 31st March 1981 while the cost was equally
shared between the central and the state government from 1982-83 (p. 32). The program envisaged
generation of gainful employment for both unemployed and underemployed workers in rural areas.
Allocation of funds from the central government was based on the incidence of poverty and the
population of agricultural laborers in a state (Annual Report MoRD 1982-83). Payment of wages
was partly in the form of food grains and partly in cash. Projects like afforestation, drinking water
wells, and community irrigation wells were undertaken under NREP. Expenditure occurred under
NREP in all the states during 1980-81 serves as exogenous introductory spending.

Rural Landless Employment Guarantee Program (RLEGP)

Implemented: 1983; Until: 1988

Program expenditure at the national level:

1983: |0.27 billion (2004 prices)
1984: |14.99 billion (2004 prices)

Exogenous change in state-level transfers:

∆si,1983 : RLEGPi,1983
∆si,1984 : RLEGPi,1984 − RLEGPi,1983

RLEGP was launched in different states/union territories on 15th October 1983. Although several
anti-poverty programs were in place for the poor at the time, RLEGP was specifically launched
4 Transfer Multipliers in India

to address the problem of unemployment for the landless laborers during the lean agricultural
season. As the Annual Report 1983-84 of the Ministry of Rural Development states “the hardcore
of rural poverty, particularly pertaining to employment opportunities for the landless during the
lean agricultural periods . . . has to be tackled in a more direct and specific manner” (p. 41). The
program’s objectives were: (i) to provide up to 100 days of employment per year to at least one
member of every landless labor household and (ii) to create durable assets to strengthen rural
infrastructure. The program was fully funded by the central government.
Construction of rural link road, land development, and soil and water conservation projects
(among others) were allowed under RLEGP with the restriction that “the wage component in a
project should . . . not be less than 50% of the total cost of the project” (p. 42). Since the program
was introduced in the second half of the financial year 1983-84, it was implemented in only 13 (out
of the 25) Indian states and was later extended to cover the rest of the states by 1984-85. Dur-
ing the first year, around |1 billion were allocated for RLEGP out of which only |62 million were
spent, possibly indicating capacity and supply-side constraints. However, expansion of the program
in the following year along with delivery mechanisms set in place resulted in program expenditure
of |3.7 billion (out of the |4 billion that were made available). Since the motivation to introduce
the program was unrelated to current or prospective output fluctuations, the resulting changes in
transfer spending during the first two years of program implementation and expansion are regarded
as exogenous variation in transfer spending. As RLEGP did not substitute any older program, it
is treated as a standalone scheme where the program expenditure at the state-level during the
year of implementation is regarded as the exogenous change in transfer spending. Moreover, since
there is evidence that the program expanded substantially during 1984-85, the increase in program
expenditure in 1984 is also considered as an exogenous introductory variation.

Jawahar Rozgar Yojana (JRY)

Implemented: 1989; Until: 1998

Program expenditure at the national level:

1989: |66.34 billion (2004 prices)

Exogenous change in state-level transfers:

∆si,1989 : JRYi,1989 − N REPi,1988 − RLEGPi,1988
5 Transfer Multipliers in India

The Annual Plan 1990-91 of the Planning Commission notes:

“Experience of implementation of NREP—RLEGP over the years showed that whereas

the Government of India’s objective in starting these programs was to tackle the prob-
lems of unemployment of unskilled rural laborers, the distribution of funds by the States
in different regions and districts did not follow this logic. The funds did not invariably
go to the areas of the concentration of unemployed landless and rural labor . . . The type
of assets being created were not economically productive. The system of approval of
the projects was such that it left much to be desired” (p. 65).

To intensify the process of employment generation and address the above concerns, the two older
employment schemes NREP and RLEGP were merged into a substantially bigger program: Jawa-
har Rozgar Yojana on 1st April 1989. In line with previous employment programs, the primary
objective of JRY was the provision of employment for the rural poor with the secondary objective as
the development of productive infrastructure. The central government financed 80% of the program
cost. Under the program, assistance from the central government to the states was determined by
the “proportion of rural poor in a State/UT to the total rural poor in the country” (p. 37) Annual
Report 1990-91 MoRD. Projects allowed under JRY were similar to the ones planned under the
previous programs. Since JRY was a result of the restructuring of two previous programs NREP
and RLEGP, I do not treat it as a standalone program. However, given the significant increase
in outlay earmarked for employment generation under JRY, I consider JRY1989 − (NREP1988 +
RLEGP1988 ) as exogenous introductory spending for the year 1989-90. Since there is no evidence
to suggest that JRY was implemented in phases, only the change in employment expenditure due
to the introduction of JRY in 1989-90 is considered to be exogenous variation in transfer spending.

Employment Assurance Scheme (EAS)

Implemented: 1993; Until: 2000

Introductory variation does not contribute to the shock series

Reason: Introductory variation is arguably endogenous

EAS was implemented from 2nd October 1993 in 1778 rural blocks of the country. The program
guaranteed 100 days of employment to a maximum of two adults per family. The primary objective
6 Transfer Multipliers in India

of EAS was to provide unskilled manual work during the lean agricultural season to anyone who is
“desirous of work, but cannot find it” (Annual Report 1993-94 MoRD, p. 19). During its second
year, the program was extended to another 665 rural blocks of the country (p. 17). Expansion
of EAS to cover all the rural blocks of the country continued till 1997-98. The program cost was
shared between the central and the state government in the ratio 80:20 (later revised to 75:25).
However, I do not treat the introductory variation due to the implementation of EAS as exoge-
nous. This is because EAS was a demand-driven scheme with no fixed allocation of funds for any
district or block. Hence according to the narrative evidence, any variation in program expenditure
is likely to be due to variation in the demand for the program which is likely to be correlated with
local economic conditions. Therefore, the implementation of EAS does not contribute towards the
construction of the narrative shock series.

Indira Awaas Yojana (IAY)

Implemented: 1995; Until: Present

Program expenditure at the national level:

1995: |17.60 billion (2004 prices)

Exogenous change in state-level transfers:

∆si,1995 : IAYi,1995 − IAYi,1994

The Annual Report 1995-96 of the Ministry of Rural Development notes:

“The Indira Awaas Yojana (IAY) was launched in 1985-86 as a sub-scheme of Rural
Landless Employment Guarantee Program to provide houses free of cost to the members
of Scheduled Castes/Scheduled Tribes and freed bonded laborers in rural areas. From
1989-90, the scheme has continued as a sub-scheme of Jawahar Rozgar Yojana. In the
beginning, 6 percent of the total allocation of the JRY was earmarked for the scheme.
From 1993-94 . . . the JRY allocation for the implementation of IAY was raised from 6
percent to 10 percent . . . From 1 January 1996 the Indira Awaas Yojana has been taken
out of JRY and made an independent scheme by itself” (p. 20).

In order to effectively target the mutually exclusive objectives of rural employment and rural
housing, IAY was implemented as an independent scheme in the financial year 1995-96. Around
7 Transfer Multipliers in India

75-80% of the program cost was financed by the central government. State poverty ratio and
the shortage in rural housing were the prime determinants in the allocation of funds from the
central government to the states. As a result of being implemented as an independent scheme,
expenditure under IAY increased to around |12 billion in 1995-96, substantially higher than ≈
|5 billion that was spent under IAY as a sub-scheme of JRY in 1994-95 (Annual Report 1998-99,
Ministry of Rural Development). Hence, the exogenous introductory variation for the year 1995-96
is calculated as the difference between IAY expenditure in 1995-96 and 1994-95. The challenge in
calculating the state-wise exogenous variation due to IAY implementation is that state-wise data
for IAY is reported only from 1995-96. Expenditure under IAY is not available at the state level
when it was implemented as a sub-scheme of JRY. To overcome this constraint, I approximate
the state-wise expenditure under IAY during 1994-95 using the state-wise expenditure data under
JRY in that year and the allocation criteria that required 10% of JRY funds to be earmarked for
IAY (which was then a sub-scheme of JRY). Using this approximation, I calculate the difference in
IAY expenditure between 1995-96 and 1994-95: IAY1995 − IAY1994 as the exogenous introductory
variation in transfer spending. Since there is no evidence to suggest that Indira Awaas Yojana
underwent any expansion in terms of coverage after 1995-96, exogenous variation is limited to the
year IAY was introduced as an independent scheme.

Swarnjayanti Gram Swarozgar Yojana (SGSY) &

Jawahar Gram Samridhi Yojana (JGSY)

SGSY Implemented: 1999; Until: Present

JGSY Implemented: 1999; Until: 2001

Program expenditure at the national level:

1999: |11.39 billion in SGSY (2004 prices)
1999: |24.19 billion in JGSY (2004 prices)

Exogenous change in state-level transfers:

∆si,1999 : (SGSYi,1999 − IRDPi,1998 ) + (JGSYi,1999 − JRYi,1998 )

Integrated Rural Development Program (that was implemented in 1978) also enveloped several sub-
schemes like Training of Rural Youth for Self Employment (TRYSEM), Development of Women
and Children in Rural Areas (DWCRA) among others. While all these schemes were supposed to
8 Transfer Multipliers in India

operate complementary to each other, over the years individual program targets replaced the larger
objective of poverty alleviation (p.40 Annual Report 1999-2000 MoRD). To address the deficiencies
identified in the older schemes, a comprehensive credit-cum-subsidy scheme: Swarnjayanti Gram
Swarozgar Yojana (SGSY) was launched on April 1, 1999, while IRDP and its allied programs
ceased to be in operation.

The focus of SGSY was to create self-employment among the rural poor. To achieve this, a multi-
pronged approach of micro-enterprise development, capacity building of the poor (self-help groups),
and credit technology were followed. The subsidy component extended to individuals/groups under
SGSY was only meant as an enabling element, while a greater reliance was on the availability of
credit. As is mentioned in the Annual Report 1999-2000 MoRD, individuals under the program
were eligible to receive a subsidy of 30% of the project cost with the subsidy ceiling of |10,000. For
group projects, the subsidy was at 50% of the project cost subject to a maximum subsidy grant of
a |1,25,000.

In the same year, the older employment program: Jawahar Rozgar Yojana (JRY) was restruc-
tured into Jawahar Gram Samridhi Yojana (JGSY) on 1st April 1999. Unlike the previous programs
like JRY and EAS where the primary objective was the generation of wage employment, the over-
riding priority of JGSY was the creation of demand-driven community village infrastructure with
the secondary objective being the generation of wage employment for the unemployed. The rule of
60:40 wage-material cost ratio outlined under the previous JRY for the creation of rural infrastruc-
ture was relaxed under JGSY so as to enable the build-up of demand-driven rural infrastructure.
Development of infrastructure support for SGSY was given a priority under JGSY. However, heavy
infrastructure investment related projects like construction of bridges, secondary schools, colleges,
and roads were not permitted under JGSY (Annual Report 1999-2000 MoRD).

For both SGSY and JGSY, the program cost was divided between the Centre and the States
in the ratio 75:25. Hence in the financial year 1999-2000, two major rural development programs:
IRDP and JRY were restructured into SGSY and JGSY respectively. The exogenous measure of
transfer spending for the year 1999-2000 accordingly measures the change in program expenditure
because of the implementation of the new restructured programs. Thus, for 1999-2000, (SGSY1999
− IRDP1998 ) + (JGSY1999 − JRY1998 ) is the measure of exogenous introductory spending. Since
there is no evidence that either SGSY or JGSY was implemented in phases, the introductory varia-
tion is limited to the year of introduction of the two programs. Since IRDP along with many other
small duplicate schemes were discontinued and replaced with SGSY, this generally resulted in an
exogenous contraction of transfer spending during 1999-2000.
9 Transfer Multipliers in India

Sampoorna Grameen Rozgar Yojana (SGRY)

Implemented: 2001; Until: 2007

Program expenditure at the national level:

2001: |47.02 billion (2004 prices)

Exogenous change in state-level transfers:

prov prov
∆si,2001 : (SGRY -Ii,2001 − SGRY -Ii,2001 ) + (SGRY -IIi,2001 − SGRY -IIi,2001 )

SGRY was launched on 25th September 2001. After the restructuring of JRY into JGSY in 1999,
EAS became the major wage-employment generation program while JGSY was more focused on
the creation of rural infrastructure. To achieve comprehensive rural development, an ambitious
program: SGRY was launched to take care of food security, employment generation, and rural
infrastructure development. SGRY initially operated under two streams where EAS and JGSY
were respectively restructured into SGRY-I and SGRY-II. Total SGRY funds were equally divided
between the two streams. SGRY was implemented as one from 2004-05 (Annual Report 2005-
06 MoRD, p. 2). Rural infrastructure development under SGRY involved projects like drought
proofing (for e.g., soil and moisture conservation), afforestation, and promotion of traditional water
resources, among others. A special component of SGRY was geared towards augmenting food
security which started from 1st April 2002 (Annual Report 2002-03 MoRD, p. 11).
The special component was a demand-driven sub-scheme where the central assistance was ex-
tended (only in terms of food grains) to the states in times of natural disasters like drought,
earthquake, cyclone, flood, etc. Naturally, expenditure pertaining to the special component is not
included in the shock series which may otherwise downward bias the multiplier estimates. The
central government financed 75% of the program cost. However, unlike most new programs, SGRY
was not introduced at the start of the financial year and was instead implemented during the end of
the second quarter of the financial year 2001-02. Construction of the exogenous variation becomes
a challenge in this case since the state-wise annual expenditure estimates for SGRY-I and SGRY-II
during 2001-02 also contain funds utilized under EAS and JGSY until October 2001.
As a workaround to this, I use the provisional expenditure estimates of SGRY-I and SGRY-II
during 2001-02 to estimate the funds utilized under EAS and JGSY between April-October 2001.
10 Transfer Multipliers in India

Annexures in the Annual Reports of MORD provide actual program expenditure estimates for the
previous years, while only provisional estimates are provided for the current fiscal year. Further-
more, the ‘reporting month’ (the month up to which the program expenditures have been reported)
is also mentioned in the financial report. In the Annual Report 2001-02 MORD, the provisional
estimates for both SGRY-I and SGRY-II correspond to the reporting month of October or Novem-
ber 2001 (for most of the states). This, consequently, provides expenditure estimates before the
prov prov
introduction of SGRY. Hence (SGRY -Ii,2001 − SGRY -Ii,2001 ) + (SGRY -IIi,2001 − SGRY -IIi,2001 )
represents the exogenous change in transfer spending at state-level due to the introduction of SGRY
in 2001. The introductory variation is limited to 2001 since there is no evidence that the program
was expanded after 2001-02.

National Rural Employment Guarantee Act (NREGA)

Implemented: 2006; Until: Present

Program expenditure at the national level:

2006: |78.69 billion (2004 prices)
2007: |132.30 billion (2004 prices)

Exogenous change in state-level transfers:

∆si,2006 : N REGAi,2006
∆si,2007 : N REGAi,2007 − N REGAi,2006

NREGA (later renamed as Mahatma Gandhi – NREGA) is often regarded as the largest workfare
program in the world. The program guarantees 100 days of manual labor work to every rural
household by organizing public works that are aimed at generating and maintaining village infras-
tructure. NREGA was implemented in phases where 200 districts were covered under the program’s
first phase in 2006-07. In the second phase, 130 additional districts were covered in 2007-08. The
program covered all the districts of the country by 2008-09. The program is almost completely
funded by the central government.
Since NREGA did not replace SGRY until 2008, NREGA’s first-year expenditure is regarded as
exogenous introductory variation while the increase in NREGA expenditure from 2006 to 2007 is
considered as exogenous variation in the year 2007. Although there is clear evidence that NREGA
11 Transfer Multipliers in India

expanded further in the third year, I do not construct exogenous variation after the second year of
NREGA. This is because by the third year it becomes progressively difficult to isolate exogenous
changes in program expenditure that occur specifically due to program expansion. The increase
in government expenditures across the board due to the stimulus package set-up by the Indian
government in early 2009, suggests that part of the counter-cyclical increase in social transfers
could be arguably endogenous. The exogenous introductory variation under NREGA is therefore
limited to 2006-07 and 2007-08.
Another valid concern is that given the size of NREGA, overall expenditure under other programs
may have declined during 2006-07 and 2007-08. However, I find no significant negative correlation
between NREGA expenditure and the year-on-year change in aggregate expenditure under SGSY,
SGRY, and IAY (at state-level) during 2006 and 2007. It is reassuring to note that the results
do not change even if I replace the NREGA shocks with the year-on-year variations in aggregate
expenditure: SGSY + SGRY + NREGA + IAY in 2006 and 2007. The latter choice of shocks
is, however, susceptible to endogeneity since expenditure variations in the other programs may
reflect local output fluctuations and hence is not the preferred choice of expenditure shocks due to
NREGA implementation.

Appendix B: Other Results

Cross-Border Spillover Effects

In this section, I analyze cross-border spillover effects of rural transfer spending. I test whether
there are any cross-border effects of spending under these programs. For example, an increase in
transfer spending in a state may spur economic activity in adjacent states due to demand leakages.
On the other hand, if higher spending under such programs results in the relocation of factors
of production from the neighboring states, then an increase in agricultural growth rate in one
state may negatively affect the agriculture sector in an adjacent state. If either type of spillover
is empirically relevant, then the baseline results will confound the local output effects of transfer
spending with demand leakage or relocation effect. Following Acconcia, Corsetti, and Simonelli
(2014), I test for cross-border effects by controlling for regional spending in the following model

3
X 3
X 1
X
(B1) Yi,t = βj Si,t−j + ξj Yi,t−j + ζj SRi,t−j + αi + γt + ri,t + εi,t
j=0 j=1 j=0

which is the same as equation 1 of the main article but with SRi,t and its lags as additional
controls. I define SRi,t = ∆sri,t /syi,t−1 where ∆sri,t is the aggregate year-on-year exogenous
12 Transfer Multipliers in India

change in real per capita transfer spending in the states that belong to the same region as the state
i, excluding the state i itself; syi,t−1 is the lagged real per capita agricultural output of the region
defined accordingly.

SRi,t in equation B1 measures whether higher spending in a region has any direct material
impact on agricultural output growth of state i, excluding program expenditure for state i itself.
Estimation of equation B1 yields coefficients of SRi,t and its lag to be −0.73 and −1.12 respectively
where both coefficients are insignificant from zero (see column 2, table B1). Hence, direct spillover
effects due to regional spending appear to be economically insignificant. The impact and cumulative
multipliers reported in column 2 are slightly higher but always comparable to the baseline results
in column 1.

In column 3 of table B1, I add the interaction term Si,t ×SRi,t (where both variables are expressed
in deviations from their respective mean) as an additional control in equation B1. The interaction
term tests whether spending in adjacent states increases (or decreases) the efficacy of program
spending at the state level. The coefficients for SRi,t and its lag at −0.92 and −1.12 respectively
are again insignificant from zero (column 3, table B1). Importantly, the coefficient on the interaction
term, estimated at 0.1, is also statistically indistinguishable from zero (t < 0.3). Therefore, the
impact of program spending at the state level appears to be quite independent of the level of
regional spending.

Overall, the multipliers reported in column 2 and 3 of table B1 are slightly higher but always
comparable to the baseline results. This may indirectly suggest weak negative spillover effects
which, when not explicitly controlled for, lead to slightly lower baseline results. However, since the
coefficients on regional spending and the interaction term are themselves not significant, I continue
to prefer the baseline estimates.

Finally, column 4 shows the results computed from the estimates of equation 1 (of the main
article) but after aggregating small states with adjacent states, thereby reducing the total number
of states from 23 to 16. This increases the impact multiplier to around 1.7 but the overall impact
after five years is 2.3, comparable to the corresponding baseline result of 2.1. In the subsection
“Effect on Aggregate State Output” below, I also check but do not find evidence for any significant
spillover effects to other sectors within a state (table B8). Altogether, the evidence of weak spillover
effects is consistent with very low levels of rural-urban labor mobility that has been identified in the
migration literature in India. Munshi and Rosenzweig (2016), for example, find internal migration
in India to be very low “both in absolute terms as well as relative to other countries of comparable
size and level of economic development”.
13 Transfer Multipliers in India

Mediation analysis

To measure the productivity mechanism, I conduct a mediation analysis using the growth in crop-
ping intensity CIi,t as the mediator variable. Cropping intensity, defined as the ratio of gross
cropped area to the net sown area, is a measure of multiple cropping on a piece of arable land in
one agricultural year. Higher cropping intensity is thus associated with higher productivity per unit
of arable land. As row 1 of table B2 illustrates, one percentage point increase in cropping intensity
increases agricultural output growth by 0.24 percentage points, thus confirming the potency of the
channel. I do not add lags of CIi,t (results do not change if I do) since there is no a priori reason
to believe that cropping intensity affects agricultural output growth with a lag. Next, to check the
effect of program spending on cropping intensity I regress CIi,t on Si,t and its three lags.

As row 2 shows, while spending shocks appear to have a small and positive effect on cropping
intensity contemporaneously and a small and negative impact at lag 1, both coefficients are in-
significant from zero. However, a spending increase of 1% of state agricultural output increases the
state’s cropping intensity by 4.9 percentage points after two years (significant at 5 percent level).
The finding of productivity effect working with a lag is consistent with public works increasing agri-
cultural productivity only upon completion of the projects. Kareemulla et al. (2009), for example,
report that although ponds and water reservoirs were built under NREGA, the connecting channels
to plots of land were constructed only later on. It is worth noting that the broader measure of
changes in transfer spending Bi,t has no effect on cropping intensity. All coefficients of spending
are small and insignificant from zero in a regression of CIi,t on Bi,t (row 3). Consistent with the
reduced form comparison (in the main article), the coefficient of Bi,t−2 in row 3 is roughly an order
of magnitude smaller than the corresponding estimate of Si,t−2 in row 2. Finally, I compare the
cumulative effect of spending shocks on agricultural output without and with the mediator variable
as an additional control in rows 4 and 5 respectively. Relative to row 4, the multiplier estimates in
row 5 are significantly lower from second year onwards. The multiplier of 0.7 in year 2 (row 5), for
example, is consistent with the finding that a multiplier of ≈ 1.2 (0.24 × 4.9) is explained through
the spending shock’s effect on agricultural output by increasing cropping intensity. Importantly,
the multiplier estimates in row 5 continue to be 30-40% lower relative to corresponding estimates
in row 4 after year 2 as well. Hence, a significant proportion of the multiplicative effect of program
spending is attributable to improved agricultural productivity from the second year onwards.
14 Transfer Multipliers in India

Effect of Spending on Employment Schemes

In the main article, I analyze the multiplicative output effects of spending under all major rural
welfare programs in India between 1980 and 2010. However, from a policy perspective, it is im-
portant to check for heterogeneity in the output response to different kinds of welfare programs.
For example, while workfare programs can affect the agricultural labor market directly, credit and
subsidy programs may have a more pronounced effect on other factors of production. While I
cannot individually estimate multipliers corresponding to credit and subsidy programs as there are
only two spending shocks corresponding to such programs (IAY in 1995 and SGSY in 1999), I can
estimate the multiplier effects due to public workfare programs as these schemes form the majority
of the programs I study. I refer to spending on workfare programs as employment spending. Overall,
I find the multiplier effects of public workfare programs to be comparable to the baseline results
including all programs.

Figure B3 and figure B4 respectively show: (i) the broader measure of year-on-year changes in
total employment spending, and (ii) the exogenous changes in employment spending, both taken
as a percent of lagged state agricultural output. Both measures are constructed exactly as in the
narrative analysis section, but only for employment schemes. For example, the narrative shock
series in figure B4 is exactly the same as the one shown in figure 2 of the main article except for
the following changes. First, the year 1995 does not have any spending innovations due to the
introduction of a housing subsidy scheme IAY. Second, the exogenous variation in 1999 is only due
to the introduction of JGSY (employment scheme) while I ignore the changes in spending that
occurred due to the introduction of a credit-cum-subsidy scheme SGSY that was also implemented
in the same year. Column 1 of table B3 shows the impact and cumulative multipliers computed
under the preferred specification with the exogenous changes in employment spending as shown in
figure B4.

The results are highly comparable to the baseline results. The impact multiplier is estimated
to be around 1.5 (t > 2.96) which grows to around 2.2 (significant at standard confidence levels)
after five years. Column 2 of table B3 shows the results from the broader measure of changes in
employment spending. Not surprisingly and consistent with the findings in the results section in
the main article, multiplier estimates obtained from this broader measure of changes in employment
spending are substantially downward biased and insignificantly different from zero.
15 Transfer Multipliers in India

Robustness

While outliers and high leverage observations can be valid observations, it is important to check
the sensitivity of the results to such observations. For example, all large shocks (greater than
two standard deviations from the mean) are NREGA shocks. A possible robustness check can be
to estimate the model after truncating the data from 2006. This removes all NREGA shocks in
the year 2006 and 2007. However, since the identification in this article is based on expenditure
shocks emanating from program introductions, removing NREGA shocks effectively removes nearly
a decade of the data as the previous program introduction was in 2001. The shock series in this
scenario contains only zero entries post-2001. Table B4 compares the impact multiplier obtained
from the baseline result with cases where I truncate NREGA shocks. As can be seen, while the
point estimate is between 1.1-1.2 for more parsimonious specifications with 1 or no lag, the standard
errors are approximately 4-5 times larger than the baseline.
The substantial loss in power of the tests is expected as truncating the time series in this case
effectively removes one-third of the total observations. The loss of power can also be motivated
from the treatment-control standpoint where the power of the test is proportional to (1/N T −1/N C
) where N T and N C represent the number of observations in the treatment and control group
respectively (where it is optimal to have N T = N C ). Specific to this article, N T correspond to the
years when new programs were introduced which are substantially smaller than N C (even in full
sample). Therefore, the loss in power is expected when I further eliminate observations from the
treatment group.
The limitation of the above test notwithstanding, it is important to check the robustness of the
results to outliers, high leverage points, and other influential points including but not limited to
NREGA shocks. For example, the smallest (−44%) and the largest (139%) growth rates of the
state agricultural output respectively correspond to the financial years 1987-88 and 1988-89 for the
state of Gujarat. However, these outliers are not data errors and instead correspond to the drought
of 1987-88 in Gujarat that “affected more than 87% area of the state” and “was the worst ever
drought during 1973-74 to 2004-05” (p. 14, Roy and Hirway, 2007). An extreme drought followed
by an average monsoon in the following year can hence explain the two outliers. I conduct four
tests to check for this.
First, I screen out NREGA shocks or any other high leverage points or influential observations
with Cook’s distance > 1. The test employs Huber and bi-weights to weigh each observation.
Results based on this model yields impact multiplier to be 1.27, marginally lower than the baseline
impact estimate of 1.36 (column 2, table B5). The multiplier after 5 years is 2.14, very comparable
16 Transfer Multipliers in India

with the corresponding baseline estimate of 2.13.

Second, I test for the robustness of the results to the exclusion of influential observations based
on the DFITS statistics. This statistic measures the influence an observation has on the overall
fit of the model to identify influential observations. After screening out approximately 5% of the
observations identified as influential, I get an impact multiplier of 1.32 and an overall multiplier of
2.46, slightly higher than the 2.13 estimate in the baseline results (column 3, table B5).
Third, I explicitly remove outliers and high leverage points in the data by removing all obser-
vations where i) agricultural growth is greater than 2 standard deviations from the mean or ii)
when expenditure shocks are greater than 2 standard deviations from the mean (greater than ≈
2% of the state agricultural output). This exclusively removes all large shocks emanating from
the introduction of NREGA. The impact and overall multiplier are again very comparable with
corresponding baseline estimates although the standard errors are relatively large due to the loss of
observations (column 4, table B5). Hence, the inclusion of these observations does not materially
affect the model’s slope coefficients but only yields more efficient estimates. Overall, the three tests
reported in table B5 suggest the baseline results to be robust to outliers, high leverage points, and
other influential observations including but not limited to NREGA shocks.
Finally, I check if multiplier responses vary with the size of shocks. I conduct Monte Carlo
simulations where I randomly select 95% of the observations in the data to estimate multipliers
based on the baseline specification. This exercise tests if the size of the multiplier is sensitive to
the average size of shocks in the selected subsample. Figure B5 plots the result of the Monte Carlo
simulations based on 1000 replications. The figure plots the overall multiplier together with the
average size of the shock based on a particular subsample. As the figure shows, the multiplier size is
orthogonal to the average shock size. This is also evident by a linear fit of the graph together with
95% confidence intervals which show the slope coefficient to be insignificantly different from zero.
Further, consistent with the baseline results, more than 90% of the replications yield the overall
multiplier to be between 1.8 and 2.5. This further corroborates that, conditional on the range of
shocks studied in the article, the multiplier estimates are quite robust to the size of expenditure
shocks.

Robustness to Different Estimators

To check the sensitivity of the results to different estimation methods, I estimate the preferred
baseline specification with (Arellano-Bond) Difference GMM and (Arellano-Bover /Blundell-Bond)
System GMM estimators. The results are reported in table B6 where column 1 reports the baseline
17 Transfer Multipliers in India

results for reference. Using Difference GMM estimator (column 2), the multiplier on impact is
estimated to be around 1.4 which increases to 2.1 after five years. For system GMM estimator
(column 3), the impact multiplier is 1.3 while the overall multiplier is estimated to be approximately
2. All estimates are significant at 5 percent level. Overall, the results are very comparable to the
baseline estimates computed using dynamic panel data using fixed effects.

Public works under NREGA

While the information on the type and number of assets generated under NREGA is limited, figure
B6 provides evidence that productive works were indeed carried out under NREGA. The figure
provides information on the total number of public works (by type of work) that were completed
under NREGA between 2006-2010. As can be seen, water-related works such as micro-irrigation
works, renovation of traditional water bodies, provision of irrigation facility, and water conservation
and water harvesting form the majority of works undertaken under NREGA. Water-related works
represent approximately 50% of the total works completed under NREGA every year.

Influence of Regions, Time, and State-Specific Dummies

I now test the sensitivity of the results to the exclusion of states based on geographic locations and
to the exclusion of year and state fixed effects.
In table B7, I respectively drop the states that belong to the northern, southern, and central
India in the columns named “Drop North”, “Drop South”, and “Drop Center” as a way to test the
heterogeneity in the results across macro areas. Next, I drop the year fixed effects to understand
how the baseline results change when I allow the multiplier estimates to be influenced by national
fiscal, monetary, and agricultural policies along with aggregate cyclical fluctuations. Finally, I drop
state fixed effects to see the extent to which cross-sectional effects impinge on the results.
While dropping states from the northern and central regions yield slightly higher impact multi-
pliers of 1.5 and 1.7 respectively, excluding the southern states from the sample yield an impact
multiplier that is a bit lower at 1. However, the aggregate effect after 5 years in all three cases is
very close to the baseline estimate of 2.1 (all significant at 5 percent confidence level). In column
4, excluding the year fixed effects yield a slightly higher impact multiplier of 1.7 which reaches a
maximum of 1.8 after three years before declining to around 1.6 after five years. Dropping the state
fixed effects in the last column yield an impact multiplier of around 1.2 with cumulative multipliers
of around 2 and 1.8 after 3 and 5 years respectively (all significant at 5 percent confidence level).
Given that the results are robust to the exclusion of state fixed effects is encouraging and lends
18 Transfer Multipliers in India

support to the identification strategy that assumes the narrative shock series to be largely orthogo-
nal to state characteristics. In conclusion, none of the test specifications yield multiplier estimates
that are substantially different from the baseline results.

Effect on Aggregate State Output

Finally, I test whether spending under rural welfare programs has any significant effect on the
aggregate state output. This is also a test to check if there are any significant demand leakages
or relocation of factors of production to other sectors within a state. Such an inquiry, however,
has some key limitations. First, the spending shocks are very small relative to state GDP which
leads to high standard errors and low power of tests. To expand on this point further, I estimate
equation B2 which is the same as equation 1 of the main article except that the dependent variable:
s is now the growth rate of real per capita state output while the transfer shock series S s is the
Yi,t
year-on-year exogenous change in real per capita transfer spending due to program introductions
normalized by the lagged real per capita state GDP.

3
X 3
X
s s s
(B2) Yi,t = βj Si,t−j + ξj Yi,t−j + αi + γt + ri,t + εi,t
j=0 j=1

Normalizing exogenous spending changes by state GDP significantly reduces the size of the spend-
ing shocks. The mean of the transfer shock series normalized by the state output is just 0.03 percent
of state GDP with a standard deviation of 0.13 percent of state GDP. Most of the spending shocks
normalized by the state GDP are on average 3-4 times smaller than the respective shocks normal-
ized by the state agricultural output. As a result of this aggregation, I expect a loss of power in
determining the true sign and magnitude of the effect that transfer spending will have on state
output. Table B8 reports the state output multipliers obtained through the narrative shock series
in column 1 and multipliers obtained through the broader measure of changes in transfer spending
in column 2 (which, like in the main article, suffer from a substantial downward bias). Importantly,
the multiplier estimates obtained from state GDP normalized shock series (in column 1) are also
insignificantly different from zero as the standard errors corresponding to the multiplier estimates
are approximately 2-3 times larger than those obtained in the baseline results.
Apart from the size of shocks, it is also important to note that the state GDP measures suffer from
substantial measurement issues which can also lead to inefficient estimates. This is also highlighted
by figure B7 which shows how the confidence bands of the state output multipliers almost always
19 Transfer Multipliers in India

envelope the confidence interval of the state-sector multipliers. The above limitations aside, such
aggregation is relevant only if fluctuations in rural transfers are representative of movements in
statewide transfers. While outside the scope of this study, future work can focus on developing
measures for state-level transfer spending that can provide better signal to noise ratio and multiplier
estimates at the aggregate state-level.
Overall, while there appears to be no significant effect of spending innovations on the aggregate
state output, the confidence intervals in this test are large enough to accommodate the baseline
results. Hence, given the limitations noted above, I do not find any concrete evidence of within
state (positive or negative) sectoral spillover effects.
20 Transfer Multipliers in India

Table B1—Cross Border Effects

(1) (2) (3) (4)

Baseline Spill-over1 Spill-over2 Aggregate
M (t) 1.361 1.605 1.489 1.630
(0.521) (0.538) (0.819) (0.722)

M (t + 1) 1.352 1.803 1.748 1.514

(0.803) (0.730) (0.768) (0.999)

M (t + 2) 2.009 2.393 2.325 2.175

(0.929) (0.940) (0.984) (1.018)

M (t + 3) 2.486 2.810 2.763 2.621

(1.152) (1.150) (1.165) (1.117)

M (t + 4) 2.180 2.501 2.435 2.306

(0.886) (0.902) (0.949) (0.866)

M (t + 5) 2.127 2.470 2.412 2.232

(0.900) (0.905) (0.938) (0.909)

-0.728 -0.916
SRi,t No No
(0.859) (1.213)
-1.178 -1.177
SRi,t−1 No No
(1.233) (1.234)
0.103
SRi,t × Si,t No No No
(0.391)
Lags of Y Yes Yes Yes Yes
Poverty Ratios Yes Yes Yes Yes
Observations 621 621 621 432
Note: The dependent variable in all the regressions is the year-on-year percentage
change in real per capita agricultural output. Column 1 shows the baseline results.
Multipliers reported in column 2 and 3 are computed from estimates of equation B1.
Column 3 additionally controls for the interaction of SRi,t with Si,t , where both the
variables are measured in deviation from the mean. Here SRi,t denotes the transfer
spending shock in the states that are in the same region as state i, excluding state i
itself. Column 4 multipliers are computed from estimates of equation 1 of the main
article but where small states are aggregated into big states. All regressions include
state and year fixed effects. The standard errors reported in parentheses are clustered
at the region × year level and are robust to heteroskedasticity.
21 Transfer Multipliers in India

Table B2—Mechanism: Mediation Analysis

CIi,t Obs.
Yi,t 0.241
460
(0.101)
Si,t Si,t−1 Si,t−2 Si,t−3 Obs.
CIi,t 0.473 -0.777 4.908 -0.208
391
(0.472) (0.675) (2.326) (0.792)
Bi,t Bi,t−1 Bi,t−2 Bi,t−3 Obs.
CIi,t
0.294 0.435 0.570 0.109
391
(0.342) (0.428) (0.614) (0.271)
M (t) M (t + 1) M (t + 2) M (t + 3) M (t + 4) M (t + 5) Ci,t Obs.
Yi,t
1.361 1.352 2.009 2.486 2.180 2.127
No 621
(0.521) (0.803) (0.929) (1.152) (0.886) (0.900)
M (t) M (t + 1) M (t + 2) M (t + 3) M (t + 4) M (t + 5) Ci,t Obs.
Yi,t
1.062 1.579 0.704 1.736 1.279 1.531 0.267
460
(0.547) (0.851) (1.152) (1.297) (1.073) (1.066) (0.106)
Note: Row 1 reports the regression of year-on-year percentage change in real per capita agricultural output Yi,t on
the year-on-year percentage change in cropping intensity CIi,t . Row 2 represents the regression of cropping intensity
on the narrative shock series Si,t and its lags. Row 3 represents the regression of growth in cropping intensity on the
broader measure of year-on-year changes in transfer spending Bi,t and its lags. Row 4 shows the multiplier estimates
that are computed based on the regression estimates of equation 1 of the main article that do not include the medi-
ator variable CIi,t . Row 5 reports the multiplier estimates that are computed based on the regression estimates of
equation 1 of the main article that in addition control for CIi,t as well. For convenience, we report the coefficient of
CIi,t in Row 5 as well. Data on cropping intensity is available only since the year 1990. All regressions include three
lags of the dependent variable, state and year fixed effects, and state poverty ratios. The standard errors reported in
parentheses are clustered at the region × year level and are robust to heteroskedasticity.
22 Transfer Multipliers in India

Table B3—Cumulative Multipliers from Spending under EGS

(1) (2)
Spending in EGS Spending in EGS
Exogenous Changes All Changes
M (t) 1.474 0.014
(0.497) (0.321)

M (t + 1) 1.348 0.115
(0.808) (0.424)

M (t + 2) 1.951 0.267
(0.939) (0.496)

M (t + 3) 2.604 0.424
(1.133) (1.116)

M (t + 4) 2.250 0.314
(0.881) (0.669)

M (t + 5) 2.205 0.314
(0.891) (0.742)

Poverty Ratios Yes Yes

Lags of Y Yes Yes
Observations 621 621
Note: The dependent variable in all the regressions is the year-on-year per-
centage change in per capita real agricultural output. Multipliers reported in
column 1 are computed using the estimates of equation 1 of the main arti-
cle but with exogenous changes due to employment programs. Column 2 re-
ports multipliers computed using estimates obtained by replacing exogenous
changes with the broader measure of all changes in employment spending in
equation 1. All regressions include state and year fixed effects. The standard
errors reported in parentheses are clustered at the region × year level and are
robust to heteroskedasticity.
23 Transfer Multipliers in India

Table B4—Multipliers after Dropping NREGA

(1) (2) (3) (4)

Baseline No NREGA No NREGA No NREGA
(3 lags) (3 lags) (1 lag) (no lag)
M (t) 1.361 0.941 1.178 1.090
(0.521) (2.327) (2.240) (2.396)

Observations 621 506 552 575

Note: The dependent variable in all the regressions is the year-on-year percentage change
in real per capita agricultural output. Columns 1-4 compare the impact multiplier under
various specifications. Column 1 reports the impact multiplier using the baseline specifica-
tion using 3 lags with data until 2010. Columns 2-4 report impact multiplier by removing
NREGA shocks in 2006 and 2007 by restricting the data till the year 2005. Column 2 em-
ploys 3 lags each of the dependent variable and the shock series. Column 3 and 4 estimate
more parsimonious specifications with 1 lag and no lag respectively. The table focuses on
the comparison of impact multiplier across specifications as the specification with no lags,
for example, does not allow computation of cumulative multipliers. All regressions include
state and year fixed effects. The standard errors reported in parentheses are clustered at
the region × year level and are robust to heteroskedasticity.
24 Transfer Multipliers in India

Table B5—Robustness to Outliers and Influential Observations

(1) (2) (3) (4)

Baseline Robust DFITS Drop
M (t) 1.361 1.265 1.316 1.230
(0.521) (0.425) (0.481) (1.173)

M (t + 1) 1.352 1.318 1.467 1.403

(0.803) (0.676) (0.729) (1.222)

M (t + 2) 2.009 2.155 2.412 2.128

(0.929) (0.752) (0.826) (1.290)

M (t + 3) 2.486 2.380 2.853 2.312

(1.152) (0.855) (0.988) (1.358)

M (t + 4) 2.180 2.236 2.569 2.226

(0.886) (0.744) (0.813) (1.298)

M (t + 5) 2.127 2.135 2.455 2.169

(0.900) (0.731) (0.807) (1.292)

Observations 621 609 591 589

Note: The dependent variable in all the regressions is the year-on-year
percentage change in real per capita agricultural output. All regres-
sions include state and year fixed effects. The standard errors reported
in parentheses are clustered at the region × year level and are robust
to heteroskedasticity.
25 Transfer Multipliers in India

Table B6—Robustness to Different Estimators

(1) (2) (3)

Baseline Diff. GMM Sys. GMM
M (t) 1.361 1.406 1.334
(0.521) (0.605) (0.641)

M (t + 1) 1.352 1.370 1.263

(0.803) (0.935) (0.907)

M (t + 2) 2.009 1.994 1.835

(0.929) (0.922) (0.922)

M (t + 3) 2.486 2.393 2.199

(1.152) (0.844) (0.780)

M (t + 4) 2.180 2.127 2.010

(0.886) (0.738) (0.743)

M (t + 5) 2.127 2.069 1.973

(0.900) (0.755) (0.742)

Observations 621 598 621

Note: The dependent variable in all the regressions is the year-on-year
percentage change in real per capita agricultural output. All regressions
include year fixed effects. Column 1 reports multiplier estimates obtained
from dynamic panel data estimation with state fixed effects. Column
2 and 3 report multiplier estimates with the (Arellano-Bond) Difference
GMM and System GMM estimators respectively. The standard errors re-
ported in parentheses are robust to heteroskedasticity.
26 Transfer Multipliers in India

Table B7—Multipliers after Dropping Major Regions and Fixed Effects

(1) (2) (3) (4) (5)

Drop Drop Drop Drop Drop
North South Center γt αi
M (t) 1.534 1.035 1.716 1.688 1.152
(0.536) (0.607) (0.576) (0.467) (0.534)

M (t + 1) 1.575 1.289 2.008 0.865 1.102

(0.892) (0.848) (0.739) (0.655) (0.734)

M (t + 2) 2.274 1.849 2.670 1.241 1.655

(1.048) (1.027) (0.906) (0.662) (0.844)

M (t + 3) 2.386 2.524 2.196 1.775 2.044

(1.268) (1.237) (1.304) (0.818) (0.876)

M (t + 4) 2.234 2.077 2.277 1.657 1.843

(0.988) (0.960) (1.102) (0.641) (0.731)

M (t + 5) 2.153 2.083 2.220 1.553 1.811

(1.002) (0.970) (1.111) (0.674) (0.731)

Poverty Ratios Yes Yes Yes Yes Yes

Lags of Y Yes Yes Yes Yes Yes
Observations 513 513 459 621 621
Note: The dependent variable in all the regressions is the year-on-year percent-
age change in real per capita agricultural output. Northern region: Haryana, Hi-
machal Pradesh, Punjab, Uttarakhand, and Uttar Pradesh; Southern region: Andhra
Pradesh, Karnataka, Kerala, Tamil Nadu; Central: Chhattisgarh, Goa, Gujarat,
Maharashtra, Madhya Pradesh, Orissa, Rajasthan. Regressions in column 1-3 in-
clude state and year fixed effects. Column 4 drops the year effects while column 5
drops state fixed effects (from equation 1 of the main article). The standard errors
reported in parentheses are clustered at the region × year level and are robust to
heteroskedasticity.
27 Transfer Multipliers in India

Table B8—Cumulative Multipliers for Aggregate State Output

(1) (2)
Shock as % of All Changes as %
State GDP of State GDP
M (t) -0.147 -0.230
(1.687) (0.513)

M (t + 1) 0.281 0.556
(1.680) (0.666)

M (t + 2) 2.465 -0.541
(1.950) (0.746)

M (t + 3) 1.952 -1.506
(2.516) (1.394)

M (t + 4) 1.960 -1.335
(2.149) (1.116)

M (t + 5) 1.686 -1.202
(2.174) (1.109)

Poverty Ratios Yes Yes

Lags of Ystate Yes Yes
Observations 621 621
Note: The dependent variable Yi,t s in all the regressions is the year-

on-year percentage change in real per capita state output. Multipliers

reported in column 1 are computed using the estimates of equation
B2. Column 2 reports the computed multipliers using estimates ob-
tained by replacing exogenous changes in transfer spending with the
broader measure of all changes in transfer spending in equation B2.
All regressions include state and year fixed effects. The standard er-
rors reported in parentheses are clustered at the region × year level
and are robust to heteroskedasticity.
28 Transfer Multipliers in India

20
% of State Ag. Output

16
12
8
4
0
-4
-8
23
21
19 2010
17
15 2005
13 2000
11
9 1995
7 1990
5 1985
3
States 1 1980 Year
Figure B1. Aggregate annual variations in rural transfer spending
1
.5
εt
0
−.5
−1

−1 −.5 0 .5 1
εt−1

Figure B2. Serial correlation of narrative shocks

29 Transfer Multipliers in India

20
% of State Ag. Output

16
12
8
4
0
-4
-8
23
21
19 2010
17
15 2005
13 2000
11
9 1995
7 1990
5 1985
3
States 1 1980 Year
Figure B3. Year-on-year changes in total employment spending

8
% of State Ag. Output

-2

-4
23
21
19 2010
17
15 2005
13 2000
11
9 1995
7 1990
5 1985
3
States 1 1980 Year
Figure B4. Exogenous changes in employment spending Programs
30 Transfer Multipliers in India

3
2.5
M(t+5)
2
1.5

.44 .46 .48 .5 .52

µ(X)

Figure B5. Monte Carlo simulation results

2006
600
2007
2008
2009
Number of Completed Works

2010
(in thousands)

400

200

0 Other Flood control Drought Micro− Renovation Provision of Land Rural Water
Works & protection proofing irrigation of traditional irrigation development connectivity conservation/
works water bodies facility harvesting

Figure B6. Number of public works (by type) completed under National Rural Em-
ployment Guarantee Act (NREGA)
31 Transfer Multipliers in India

2
Percent

-1

-2 Exogenous Changes as a % of State Agri GDP

Exogenous Changes as a % of State GDP
-3
0 1 2 3 4 5
Year
Figure B7. Impact of transfer spending on state GDP vs. state agricultural GDP
32 Transfer Multipliers in India

REFERENCES

Acconcia, A., G. Corsetti, and S. Simonelli (2014). Mafia and Public Spending: Evidence on the
Fiscal Multiplier from a Quasi-experiment. American Economic Review 104 (7), 2185–2209.

Kareemulla, K., P. Ramasundaram, S. Kumar, and C. A. Rama Rao (2009). Soil and Water Con-
servation Works through National Rural Employment Guarantee Scheme (NREGS) in Andhra
Pradesh — An Analysis of Livelihood Impact. Agricultural Economics Research Review 22,
443–450.

Ministry of Rural Development (Various issues). Annual reports, Government of India.

Munshi, K. and M. Rosenzweig (2016). Networks and Misallocation: Insurance, Migration, and
the Rural-Urban Wage Gap. American Economic Review 106 (1), 46–98.

Planning Commission (Various issues). Annual Plan, Government of India.

Roy, A. K. and I. Hirway (2007). Multiple Impacts of Droughts and Assessment of Drought Policy
in Major Drought Prone States in India. Gujarat, India: Centre for Development Alternatives.