Does the Quality of Electricity Matter?
Evidence from Rural India
Ujjayant Chakravorty
†
Martino Pelli‡ Beyza Ural Marchand§
April 2013
Preliminary, please do not cite
Abstract
This paper estimates the returns from increased access to electricity in rural India
on household income. We examine the effect of connecting a household to the grid
(the extensive margin) and the quality of electricity, in terms of daily hours of supply
(the intensive margin). The analysis is based on two rounds of a representative
panel of more than 10,000 households. We use the district-level variation in land
elevation and the district-level density of transmission cables as instruments for the
electrification status of the household. We find that a grid connection increases nonagricultural incomes of rural households by about 20-30 percent during the study
period (1994-2005). However, a grid connection and a higher quality of electricity (in
terms of fewer outages and more hours per day) increases non-agricultural incomes
by about 85 percent in the same period. This is equivalent to an income growth of
about 8 percent per year. Moreover, marginal returns from average daily hours of
supply are highest at very low and very high levels of daily supply, not over a wide
range in-between.
JEL classification: O12, O18, Q48
Keywords: Electricity Supply, Quality, India, Energy and Development, Infrastructure
†
Department of Economics, Tufts University, 114A Braker Hall, 8 Upper Campus Rd., Medford, MA
02155-6722, USA. Email: Ujjayant.Chakravorty@tufts.edu
‡
Department of Economics and Alberta School of Business, University of Alberta, BUS 3-40E, Edmonton, AB, T6G 2R6, Canada. Email: Martino.Pelli@ualberta.ca
§
Department of Economics, University of Alberta, Tory 7-12, Edmonton, AB, T6G 2H4, Canada.
Email: Beyza.Ural@ualberta.ca
1
1
Introduction
In 2009, 288 million people – about a fourth of India’s population – had no access
to electricity (World Energy Outlook 2011). In 2005, 364 million people did not have
access, while another 300 million received intermittent supply (Balachandra, 2011). The
massive grid failure of July 2012 affected 670 million individuals, almost 10 percent of
the world’s population and made world headlines (New York Times, 2012). According
to the World Bank, unreliable electricity supply has been a major obstacle to Indian
economic development, limiting its comparative advantage in labor-intensive products
(World Bank, 2010; Rud, 2012). Connecting all households to the grid and providing
them assured electricity is likely to have a major impact on the Indian economy.
The aim of this paper is to estimate the economic returns to rural households from
a grid connection (the extensive margin) as well as from the quality of electricity supply
(the intensive margin).1 We study the effect on rural household income that includes
all income from wages and business income, except income from agricultural assets for
which we have insufficient data.2
Rural electrification may affect incomes of households in several ways (Oda and Tsujita, 2011; Modi, 2005). It may free members of the household from domestic chores,
or let them perform these tasks in the evening. The resulting increase in the labor supply may drive down wages, especially for females. Electrification may also increase the
productivity of some activities. The productivity of agricultural labor may improve due
to technologies such as sprinkler or drip irrigation, which is likely to have an upward
effect on wages. Assured supply of electricity creates opportunities for entrepreneurial
activities which can take place within the household, such as rice-milling and production
of oil from oilseeds. Other ancillary industries, such as the repair and welding of agriculture implements such as ploughs and tractors, may now be possible. These activities are
likely to increase farm incomes, which may result in a general increase in labor supply
and wages. Finally, electrification could also affect labor supply by children. The need to
collect different kinds of fuels, including animal and agricultural waste and fuelwood may
diminish and therefore their labor supply to market activities may increase, or they may
allocate their time to other pursuits such as education.The possibility of switching from
biomass to electricity for cooking may improve indoor air quality and have a positive
impact on health (Modi, 2005; Oda and Tsujita, 2011).
There are many studies that qualitatively describe the Indian electricity sector (Balachandra, 2011; Khandker et al., 2010; Bhide and Monroy, 2011). Most papers identifying
the effect of rural electrification using econometric techniques such as propensity score
1
2
We define “quality” in terms of average hours of daily electricity received by the household.
70 percent of India’s population lives in rural areas, the focus of our analysis.
2
matching, difference-in-difference or instrumental variables usually deal with countries
other than India. Bensch et al. (2010) focus on the case of Rwanda, while Khandker et
al. (2009a and 2009b) study Bangladesh and Vietnam, specifically rural electrification
initiatives undertaken by the World Bank. Two recent papers focus on the extensive
margin of electrification. Dinkelman (2012) studies the labor market effects of an electrification project in South Africa using an instrumental variable approach. Lipscomb et
al. (2013) focus on the case of Brazil and, using instrumental variables, find a positive
impact of electrification on various development indicators. The main contribution of
our paper is the focus on the intensive margin of electricity, which has not been done
previously.
A variety of factors may lie behind the electrification of certain areas and the nonelectrification of others. Governments usually aim infrastructure investment to already
growing areas. Other economic trends may affect the investment decision. For instance,
a rich village is probably more likely to be electrified than a poor village. The likelihood
of being connected to the grid may also depend on the proximity to a big city, or on
the population density of the region. For all these reasons, disentangling the impact
of infrastructure investments such as electricity on development outcomes has ben discussed extensively in the literature (Duflo and Pande, 2007; Roller and Waverman, 2001;
Aschauer, 1989; Garcia-Mila and McGuire, 1992; Holtz-Eakin, 1993).
We tackle this endogeneity issue regarding electrification using an instrumental variable approach. We use a set of two instruments. First, following Dinkelman (2012) we
use the district-level variation in land elevation. This variable, measured as the standard
deviation of the mean elevation strongly correlates with the cost of electrification. Setting up a transmission network on flat land is cheaper than installing it on undulating or
hilly terrain. A high variability in land elevation implies a larger number of structures
(e.g., poles) must be put in place in order to ensure that the cables do not sag and run
too close to the terrain. There may be a trade-off between building a longer route for
the grid with higher transmission losses and a shorter route (e.g., going over the top of a
mountain) which is likely to be relatively expensive. A high degree of variability in land
elevation is expected to decrease the probability of electrification.
We construct another instrument which measures the variation in the density of
transmission cables in each district from the national average. The data on transmission
network is constructed by measuring the length of the transmission cables within each
district using ArcGIS. The argument is straightforward. If a household is located in
a district served by a higher density of transmission cables the probability of being
connected to the network, and by receiving a better quality power supply, is higher than
when it is located in a district with a lower cable density (Brown and Sedano, 2004). In
addition, because transmission lines are connected to many power plants, a household in
3
a district with higher density of transmission cables are more likely to receive reliable,
high quality power.
The decisions regarding investments on a transmission network in India lie with the
federal government, but not the state governments (Modi, 2005; Balachandra, 2011).
Transmission lines are a major infrastructure investment and require a high voltage pole
at their destination. The primary purpose of establishing transmission lines is to electrify
urban and industrial areas (Brown and Sedano, 2004). Once the federal government decides to build a transmission line between two points, the planner’s problem is to find the
shortest and the least costly route between these two points, and the objective of bringing
electricity to remote villages is not considered during this stage.3 For this reason, the
objective of rural electrification in India played little to no role on how the transmission
lines were installed and located. In fact, the federal-level Indian legislation did not explicitly mention rural electrification until 2003 (Modi, 2005).4 On the other hand, elected
state officials within the states have large incentives to improve rural electrification, but
they only have the control of investments in distribution networks (Balachandra, 2011).
While distribution networks are much less costly, they can only be be installed if there is
a transmission line within a feasible radius. For these reasons, this paper uses the transmission network, rather than distribution network, as an instrument for the probability
of electrification in rural areas.
We find that a grid connection increases non-agricultural incomes of rural households
by about 20-30 percent during the study period (1994-2005). However, higher quality of
electricity (in terms of fewer outages and more hours per day) increases non-agricultural
incomes by about 85 percent in the same period. This is equivalent to an income growth
of about 8 percent per year. We find that assured electricity supply increases wage
incomes, but surprisingly, reduces the income of small businesses or households engaged
in petty trades. At the intensive margin, returns are highest at low and high daily hours
of supply but not for a wide range in-between.
The remainder of the paper is organized as follows. Section 2 discusses electrification
policy in India. Section 3 describes the data used in our analysis. Section 4 summarizes
the descriptive statistics and presents stylized facts. Section 5 outlines the methodology
used and presents results. Section 6 concludes the paper.
3
The per mile cost of setting up a new transmission network may range from several hundred thousand
dollars to several million (Brown and Sedano, 2004).
4
In the federal planning budget, only about 10 percent of the electricity outlay is spent on distribution,
the rest goes towards power generation and transmission.
4
2
Background
Electricity supply in India is inefficient at all stages of production, transmission and
distribution (Modi, 2005). At subsidized prices, supply lags demand, especially during
peak periods, leading to frequent outages and voltage fluctuations. In 2007, average per
capita consumption was 543 kWh, lower than Sub-Saharan Africa (578 kWh) while the
average for OECD countries was 8,477 kWh. Average per capita consumption in India
is about a fourth of Chinese consumption (2,346 kWh).5 These low figures are partly
because India is home to a third of the world’s population that is not connected to any
electric grid. Cost recovery is low and has actually declined over time (69 percent in 200102). These low rates are mainly due to losses during transmission and distribution, which
rose from 25% in 1997-98 (Modi, 2005) to 38.86% in 2000-01 (Oda and Tsujita, 2011),
while the average for neighboring countries is about 10%. Estimates suggest that only
55% of the power supplied is billed and only 41% is paid for (Modi, 2005). Infrastructure
theft has led to further declines in coverage (Balachandra, 2011).
Figure B.1 and Figure B.2 plot the transmission network with the rail and road networks, respectively. Note that both rail and road networks have a national coverage while
the electricity transmission network is quite unevenly distributed, with higher coverage
in the northern industrial states. Large parts of the country have no transmission lines
hence no electricity. Regions with poor transmission network include eastern Karnataka,
western Andhra Pradesh, Jammu & Kashmir, Arunachal Pradesh, Assam, Aizawl, Imphal, Nagaland, Gujarat (outside of the Ahmedabad metro region), western Rajasthan
and eastern Madhya Pradesh. However, there has been significant improvement in average consumption and production of electricity. The generation capacity grew from 1,362
MW in 1947 to nearly 74,699 MW in 1991. Over the same period, per capita consumption increased from 15.55 kWh to 252.7 kWh (Modi, 2005). Currently, capacity exceeds
170 GW.
According to Brown and Sedano (2004), the cost of setting up a transmission network ranges from several hundred thousand dollars, if there are no geographical obstacles
such as hills or mountains, to over a million dollars per km. Figure 1 shows the main
components of an electricity network. The transmission network carries electricity from
production sites to high demand locations. This network is characterized by a high voltage (High-Voltage Direct-Current of 765, 500, 345 and 230 kV), as this reduces transport
losses. Because of the high voltage, it is not possible to directly connect a house or a
firm to the line, and a step-down transformer must be installed to decrease the voltage
and then channel power through a subtransmission network, which is characterized by
lower voltage, usually 69, 115 or 138 kV. At the end of the sub-transmission network
5
International Energy Agency, 2009.
5
there is another electrical substation which moves electricity into the distribution network. This last section of the grid is characterized by lower voltages (less than 50 kV)
and travels very short distances, between 50 and 100 km. The electricity transported by
the distribution network is at a voltage that can be consumed by the retail consumer.
Figure 1: Schematic of the Power Grid
Source: United States Department of Energy.
Setting up new segments of the transmission network is costly and differs from the
investment in the distribution network. The cost depends on its length, but also on
other factors related to the landscape it must traverse. For example, installing cables
over rolling hills is costlier than over flat land because the line will need a higher number of
supporting structures. If the line has to bypass a mountain, the cost is likely to increase,
as more structures are needed. Because of the high cost of setting up a transmission
network, and because it is mainly used to transfer electricity to high demand centers,
its route does not serve retail customers between the two points (Brown and Sedano,
2004). The engineer’s problem at this stage is to find the route that is shortest and the
most cost effective. Once the transmission cables are installed, the state government
can withdraw power through a step down transformer and building a new distribution
network. In our data sample, over 80% of the households live in villages with populations
less than 5,000 and out of these, 24% live in villages with less than 1,000 inhabitants.
These small villages are not likely to have any impact on the location of transmission
lines, an observation critical to our identification strategy.
6
We differentiate between connecting to the power grid (the extensive margin) and the
quality of power supply (a measure of outages and average daily hours of supply which
are generally correlated - the intensive margin). 6 The density of transmission lines thus
determines the quality of power supply, as well as the probability of being connected to
a grid. A household situated in a district characterized by a high density of transmission
cables is more likely to be connected to the grid, and conditional on being connected, it
is more likely to receive higher quality power.
3
Data and Stylized Facts
The panel contains two survey waves conducted in 1994 and 2005. The first wave is part
of the Human Development Profile of India (HDPI) and covers 33,230 rural households.
A share of these households was then re-interviewed for the India Human Development
Survey (IHDS) in 2005, which covered 41,554 rural and urban households. The households to be re-visited in 2005 were chosen by first randomly ranking villages, and then
fulfilling a variety of conditions needed for the survey to be representative by starting
in the first village and then moving down the list of villages. The survey is thus representative at the village level, but not necessarily at larger geographical units such as a
district. The final data consists of a representative panel comprising 12,958 rural households which were interviewed both in 1994 and 2005. Households are asked whether or
not they receive electricity, and if they do, information is obtained regarding the intensive margin of the electricity supply. While the definition of power quality varies across
rounds, we carry out various robustness tests to ensure that the impact of power quality is correctly identified. The panel also contains a wide variety of information at the
individual, household and village level.
Our main variables are constructed as follows. House is a dummy variable which
takes a value of 1 if the household owns the house in which the family is currently
living. Children, Teen and Adult represent the percentage composition of the household
in terms of children (between 0 and 14 years old), teens (between 15 and 21 years old)
and adults (older than 22 years old), respectively. Size represents the total number of
individuals in the household. Livestock is a dummy variable which takes the value 1
if the household owns farm animals. Hindu is a dummy variable taking a value of 1 if
it is a Hindu household and 0 if it belongs to another religious denomination. Finally,
the occupation of the household head is constructed by grouping occupational categories
under broader classes to achieve concordance between the two rounds of the survey.7
6
We use the term "quality" of power rather than "reliability" which in engineering literature has a
precise meaning relating to interruptions in power supply. This will become clearer when we discuss the
data.
7
The complete list of occupations for each round is presented in table A.1.
7
These categories are Agriculture, Low Skill and High Skill, indicating whether the head
of the household is affiliated with agricultural, low skill non-agriculture or high skill
non-agriculture occupations, respectively. The control group, Other, covers students,
unemployed individuals, and individuals engaged in home-based activities.
Finally, household income is measured as non-agricultural income in per-adult equivalent terms. It is computed by using the OECD definition of equivalence scale, which
is given by 1 + 0.7(NAdults − 1) + 0.5NChildren (OECD, 1982). This expression gives us
the number of adult equivalent members of the household. There is some discrepancy
in variable definitions between the two surveys. For example, the questionnaire used in
2005 was much more detailed on agricultural income than the one used in 1994. The
latter questionnaire also contains questions about losses, while in the former, agricultural
income was not allowed to be negative. Therefore, for the sake of our analysis, data on
agricultural income could not be included in household total income.
Table 1 presents the household characteristics considered in this paper. Per adult
equivalent income increased significantly over the period studied, and so did its volatility.
On average, there were 6.6 individuals per household, while this number was reduced to
5.5 by 2004. About one third of the individuals within households were children, 14
percent were teenagers and slightly more than half of the household population were
adults. The share of adults has increased over time, and the share of children as well as
teenagers has decreased. Approximately 85 percent of the households in the sample are
Hindus. With respect to asset ownership, almost the entire sample owned the dwelling
they were living in. In 1994, 70 percent of the sample owned livestock while this number
increased to 85 percent in 2005. The proportion of household heads employed in a skilled
occupation is very low, but increased significantly between 1994 and 2005, going from
0.3% to 1.91%. Household heads dedicated to low skilled occupations remained around
30%, while those working in agriculture diminished significantly from 56% to 30%. Many
factors may explain this shift: e.g., some of them may have just stopped working and
retired.
3.1
Household Electrification
Table 2 reports household grid connection rates for the 18 states surveyed in this study.
The majority of the states surveyed exhibit an increase in connection rates.8 The table
shows that the better-off states, such as Andhra Pradesh, Tamil Nadu and Kerala are also
those doing better in terms of grid connections. Poorer states, such as Bihar, Orissa and
8
Assam shows a 100% connection rate, but because of random sampling, only three households from
Assam were interviewed.
8
Table 1: Household Characteristics
Variable
Income per Adult Equvalent
Household Size
Share of Children
Share of Teenager
Share of Adults
Hindu
Home Ownership
Livestock Ownership
Occupation: Agriculture
Occupation: Low Skill
Occupation: High Skill
Mean
6.472.4
6.577
0.341
0.146
0.513
0.854
0.968
0.704
0.564
0.344
0.003
1994
Std. Dev.
7,639.8
3.322
0.21
0.166
0.187
0.353
0.175
0.456
0.496
0.475
0.056
Mean
8,722.5
5.546
0.3
0.136
0.564
0.848
0.98
0.845
0.299
0.276
0.019
2005
Std. Dev.
10,895.0
2.759
0.222
0.173
0.216
0.359
0.139
0.361
0.458
0.447
0.137
Obs.
9791
9791
9791
9791
9791
9791
9791
9791
9782
9782
9782
Notes: The table reports descriptive statistics of the household controls used in the
estimation. The following variables are dummies: Hindu, Home and Livestock. Child,
Teen and Adult represent percentages and finally Occupation Head Agriculture, Low
Skill and High Skill represent the percentage of household heads employed in the different occupation groups, which are described in Table (A.1).
Uttar Pradesh are under-performing.9 Figure 2 provides an illustration of the districtlevel electrification rate in 2005, which shows that for almost half of the districts in our
sample, the electrification rate is less that 57 percent, with a high level of variation across
districts.
The quality of power supply at the household level is constructed as follows. The 1994
wave asked if the household’s power supply was continuous, if the household experienced
on average one or two or more than two power outages per week. The electricity supply
is characterized as high quality if the household reported no power outages. We assume
this threshold for high quality to be 18 hours per day.10 The 2005 wave asked how
many hours of electricity per day does the household receive on average. The quality of
power supply variable is then defined as a variable that takes the value of zero when the
household is not connected to the grid, 0.5 when it is connected but receives a low quality
power supply, and 1 when it is connected and receives high quality power supply. As
shown in Table 2 the quality of supply improved by almost 80% between 1994 and 2005.
Quality increased across the board, and the numbers show a certain degree of correlation
between the increase in grid connections and quality.
9
This relationship between income and electrification rates is clearer at the state level. Figure B.3
shows that there is a strong and positive correlation between state-level income and electrification rates
in 2005. The same relationship holds for 1994, not presented here for brevity.
10
Different values of the threshold – 16 and 20 hours – were used in robustness checks, and the results
hold over a large set of thresholds.
9
Table 2: Household Electrification Rate by State (%)
Electrification
1994 2005 Change
Quality of Supply
1994 2005 Change
Andhra Pradesh
Assam
Bihar
Chhattisgarh
Gujarat
Haryana
Himachal Pradesh
Jharkhand
Kerala
Maharashtra
Madhya Pradesh
Orissa
Punjab
Rajasthan
Tamil Nadu
Uttar Pradesh
Uttarakhand
West Bengal
54
0
10
36
71
82
91
20
76
67
62
17
77
49
66
19
33
13
86
0
21
68
87
90
97
64
89
78
71
34
95
54
88
39
66
36
32
0
11
32
16
8
6
44
13
11
9
17
18
5
22
20
33
23
31
0
7
27
44
45
69
17
44
40
41
21
45
36
50
19
19
15
78
0
27
57
80
60
99
71
96
80
45
94
64
49
98
50
61
86
47
0
20
30
36
15
30
54
52
40
4
73
19
13
48
31
42
71
India
53
69
16
39
71
32
Notes: The table reports the percentage of households that reported a grid
connection in 1994 and 2005.
3.2
Infrastructure: The Transmission Network
In order to construct the variable for the length of the transmission cables within each
district, we use maps of the Indian transmission network published by the Indian Ministry
of Power. We then superimpose these maps on the map of Indian districts from Census
2001 (Government of India, 2001) using ArcGIS. Next, transmission cables are split along
the borders of the districts in order to measure their exact length within each district.
The density of the transmission network is then computed by taking the total length
of all the transmission line segments for each district, and dividing it by the district
surface area. The deviation of this measure from the national average is defined as the
normalized transmission cable density. Figure 3 shows the map of this cable density
measure for 2005.11 The shades of green represent values below the national average,
while warmer shades show above average densities. If a household is located in a district
characterized by a positive density, then the probability that the household is connected
to the grid, and receive high quality electricity is higher (Brown and Sedano, 2004), while
the opposite is true for negative values.
Table 3 shows the number of households with and without connection in the two
11
The map for 1994 is presented in Figure B.4
10
Figure 2: Electrification rate in the survey districts in 2005.
Note: Source is ESRI ArcGIS World package and geocommons. Districts left white are not part of the
survey sample.
periods. Those with electricity are divided into two groups - higher or lower than a
daily average of 18 hours of supply. During the study period, the number of households
connected to the grid increased from 52.9% initially to 68.7%. Table 3 also shows the
variation in real income per adult equivalent between 1994 and 2005. Households experiencing the highest increase in income per adult equivalent are those who got connected
to the grid within that time period. This data suggests a strong positive effect of electrification on income.12 The number of households with low quality power decreased by
21.6%, while those with good supply more than doubled. The status of a small number
of households regressed, either from good to poor power supply, from poor to no grid
connection and also from a good to no connection. The last may have occurred because of
theft of cables and other infrastructure, which is common in India (Balachandra, 2011).
Both being connected to the grid and receiving more hours of supply have a positive
effect on household income (see Table 3, Figures B.3). Table 3 shows (top row) that
12
Approximately three quarters of the households report data on the hours of electricity they received.
The full sample is used in the analysis wherever possible.
11
Figure 3: Density of transmission cables per km2 by district, 2005
Note: Source is ESRI ArcGIS World package and geocommons. The cable density is normalized as a
deviation from the national mean.
moving from no connection in 1994 to a connection in 2005, without regard to the quality
of power supply, leads to an increase in household income of about 3%. This number is
statistically significant at the 1% level. Moving from a low to a high quality power supply
(second row) generates an increase in income of 3.6 %, which is also statistically significant
at the 1% level. These preliminary figures seem to suggest that both connecting to the
grid and improving the quality of supply are important factors.
Table 4 summarizes information on the density of transmission cables by state. Between the years we study, the transmission network expanded by about 23% (more than
7,000 kms). However, there is considerable heterogeneity within states - West Bengal
saw a decline of 3.3% while the network size in Gujarat increased by 96.7%. The states
which saw their network size increase – Gujarat, Maharashtra, Punjab, Rajasthan and
Uttar Pradesh – also experienced large gains in mean household income.
12
Table 3: Household Electrification and Quality of Supply
Not Connected
2005
Connected,
Low Quality Supply
Connected,
High Quality Supply
Total
2,447
1.9∗∗∗
1,188
2.8∗∗∗
978
3.4∗∗∗
4,613
Connected, Low Quality Supply
Number of Households
Change in Income (%)
532
0.09
1,762
2.1∗∗∗
1,848
3.6∗∗∗
4,142
Connected, High Quality Supply
Number of Households
Change in Income (%)
83
2.3
299
2.1∗∗
653
4.6∗∗∗
1,035
3,062
3,249
3,479
9,790
1994
Not Connected
Number of Households
Change in Income (%)
Total
Number of Households
Notes: The table reports the absolute number of households moving from one state to the other between
1994 and 2005 and the variation in the household total income, not including income from agriculture. Asterisks report statistics for a test of two means between 1994 income and 2005 income. *** p<0.01, **
p<0.05, * p<0.1. High Quality Supply is defined as more than 18 hours of electricity per day on average in
2005; in 1994, households were asked whether or not electricity supply was "reliable."
3.3
Geography: The Land Elevation
The district level variation in land elevation is computed from Global Land Cover Facility
data (USGS, 2004), which provides maps of global land elevation at 90 meters resolution.
Each district is divided into grids of 90 by 90 meters, and the grid elevation with respect
to sea-level is known. Using ArcGIS, we compute the mean, standard error and range of
elevation for each district. While the range gives us the difference between the lowest and
highest points in a district, the standard deviation gives us the variation in the terrain.
For instance, rolling hills will generate a higher variation than a gentle upward slope. A
larger variation implies a higher cost of electrification.
Table 5 reports mean land elevation by state. Figure 4 shows the district-level variation in land elevation. The northern states closer to the Himalayas, such as Himachal
Pradesh and Uttarakhand, report the highest mean elevation and a bigger range. These
regions also exhibit large variation in elevation. Within the states with high variation
in land elevation, there are relatively poor states such as Assam, and richer states like
Kerala. The coastal districts are dropped as their measurement is not precise: the map
used assigns extreme values to the pixels close to the coast. This corresponds to dropping
8 of the 157 districts sampled.
13
14
32,432.0
2,057.2
465.5
1,635.1
368.6
1,118.5
736.9
281.7
1,320.0
1,368.0
2,337.9
2,359.0
3,568.6
1,018.0
1,076.2
3,041.5
5,326.9
323.2
807.4
0.0101
0.0074
0.0055
0.0169
0.0027
0.0060
0.0167
0.0051
0.0160
0.0362
0.0076
0.0076
0.0224
0.0202
0.0032
0.0233
0.0220
0.0060
0.0091
28,710.6
24,192.8
33,245.5
25,179.1
18,060.5
23,179.1
36,755.6
33,296.9
39,286.0
30,393.1
25,364.8
17,754.8
19,556.5
38,883.9
30,476.7
30,620.9
28,370.0
22,342.3
26,830.0
157,031.0
253,971.0
39,779.3
2,281.9
465.3
1,644.6
383.6
2,186.9
978.3
292.6
1,326.7
1,566.0
3,100.4
2,965.8
3,621.0
1,673.0
2,132.1
3,422.0
6,798.5
323.1
785.6
Total
length
0.0163
0.0258
0.0124
0.0082
0.0055
0.0170
0.0028
0.0118
0.0222
0.0053
0.0161
0.0414
0.0101
0.0096
0.0227
0.0332
0.0062
0.0262
0.0280
0.0060
0.0088
Length/km2
Real mean
income
Length/km2
22,564.0
27,215.3
10,921.8
24,842.4
16,857.7
22,626.3
52,558.4
51,985.8
45,039.8
54,139.0
28.537.8
18.673.2
19,977.2
55,157.2
34,403.1
28,513.2
35,848.6
29,570.4
28,439.0
Real mean
income
9,640,821.0
9,826,675.0
3,195,023.0
276,472.5
84,841.0
96,875.8
137,771.5
185,460.2
44,017.4
54,997.7
82,330.5
37,824.5
307,545.2
308,997.3
159,348.7
50,330.2
341,790.7
130,586.9
242,618.2
53,833.8
88,992.4
Total
surface
157
11
1
8
7
10
11
8
4
4
16
18
11
8
12
8
9
3
8
# of district
surveyed
Notes: Real mean income is calculated using the households included in the sample. For example only three households are interviewed
in Assam, causing significant variation in mean income. 1 Yang (2006). 2 Brown and Sedano (2004)
China (2000)1
US (2002)2
India
Andhra Pradesh
Assam
Bihar
Chhattisgarh
Gujarat
Haryana
Himachal Pradesh
Jharkhand
Kerala
Maharashtra
Madhya Pradesh
Orissa
Punjab
Rajasthan
Tamil Nadu
Uttar Pradesh
Uttarakhand
West Bengal
Total
length
2005
1994
Table 4: The density of transmission cables
Figure 4: District level variation in land elevation
Note: Source is Global Land Cover Facility (USGS, 2004). Darker red indicate a higher degree of variability
in district-level land elevation. The measure is not computed for coastal districts (white). Elevation for
these districts cannot be accurately measured.
4
Empirical Approach and Results
We deal with the endogenous probability of electrification and quality of supply by using
an instrumental variable approach, following the seminal work of Angrist and Imbens.13
The endogeneity of electrification applies to both the probability of receiving electricity,
and the quality of electricity supply. We use two different instruments. First, following
Dinkelman (2012), the connection to the grid is instrumented using the district level
variation in land elevation. As the variation increases, mean gradient within the district
is higher and therefore it is more costly to set up distribution lines. Second, the quality of
power supply is instrumented with the density of transmission cables within the district
relative to the average over all districts. Proximity to the transmission network implies a
closer connection to the source of electricity and therefore a higher probability of receiving
good power supply.
13
See, for example, Angrist (1990), Angrist and Imbens (1994) and Angrist (1998).
15
Table 5: Average Land Elevation (meters)
Andhra Pradesh
Assam
Bihar
Chhattisgarh
Gujarat
Haryana
Himachal Pradesh
Jharkhand
Kerala
Maharashtra
Madhya Pradesh
Orissa
Punjab
Rajasthan
Tamil Nadu
Uttar Pradesh
Uttarakhand
West Bengal
Range
Mean
St. Dev.
1,145.7
1,711.7
660.2
1,122.0
973.8
1,146.7
5,691.2
1,099.1
2,278.7
1,233.5
1,023.2
1,356.1
634.6
1,428.3
2,318.0
695.9
6,612.9
3,197.7
349.
113.7
69.0
388.7
136.9
231.6
1,914.1
291.0
469.6
415.5
378.0
279.1
220.8
279.1
322.6
134.1
1,694.9
109.4
96.6
69.5
23.0
106.8
65.5
22.7
625.3
86.7
321.2
82.0
71.8
127.1
20.6
59.4
195.0
13.4
749.1
73.6
Notes: Coastal districts are not included because elevation cannot be accurately measured.
The main specification includes the village and year fixed effects to account for variation across villages in terms of their proximity to cities, industrial regions and other
village-level characteristics such as land quality, other infrastructure including roads and
railways that may affect income levels and may be time-varying. These controls also
account for any changes in electricity prices that may affect incentives to connect to the
grid. We also estimate the main specification by excluding villages that are near a main
city, and excluding villages that are located within a district with high share of industrial
output.
At the household level, the investment necessary for connection to the grid may be
better justified for those with significant land endowments, or those engaged in some
home-based economic activity. Households that can invest in electrical appliances may
find it worthwhile to connect to the grid. Even for households connecting to the grid
illegally, the initial connection may require some expenditure in the form of a deposit
to cover fixed costs for cables and other equipment, that may rule out poorer households. We present our results with a full set of household controls for factors that relate
to the general well-being of the household, including assets, occupation and household
composition.
Let Y denote household income and X be a vector of household-level controls. Define
T to be the treatment variable, which represents one of the two electrification variables:
the extensive margin variable which is an indicator of whether the household receives
16
electricity, and the intensive margin variable, which represents the quality of power supply
received by the household. Let Z be the district-level instrument. Our specification takes
the following form:
Yit = α + δt + δv + β T̂it + Xγ + εit
(1)
Tit = ϕ + δt + δv + ηZit + Xζ + uit
(2)
where subscripts i, t and v denote household, time and village level variables, respectively.
Thus, δt and δv represent time and village fixed effects and ε and u are error terms.
The coefficient of interest is β. The first stage, regression (2) is modeled with a probit
specification when T is the indicator for receiving electricity, and linear specification when
T represents quality of power supply. All standard errors are clustered at the village level
to account for within-village correlation.
Table 6 presents first stage regressions to assess the relationship between transmission
cable density and electrification and quality of power supply in villages. The model
is estimated using various combinations of demographic controls, including household
size, share of children, teen and adult members within the household, asset controls
including livestock and home ownership, and occupational controls indicating whether
the household head works in agricultural, low skill or high skill occupations. The results
suggest that cable density has a positive and robust impact on electrification in these
remote villages. The coefficient is robust at 1.9 percent across all specifications, implying
that a one unit increase in the density of transmission lines increases electrification by
1.9 percentage points. The results also show that quality of electricity increases with
transmission cable density, with one unit increase in cable density increases the quality
index by 2.0 percentage points.14
4.1
Extensive Margin: Connecting to the Grid and Household Income
The impact of electrification on household income is presented in Table 7. The model is
presented with OLS and IV specifications, the latter using the density of transmission
cables as instrument for electrification. As before, standard errors are clustered at the
village level. Time and village fixed effects are included in all specifications to account for
time-invariant characteristics of villages, as well as changes over time that are common
to all villages.
The OLS results without household controls presented in Column (1) indicates that
14
Note that the number of observations declines in a panel due to missing information on power quality.
In order to deal with this issue, we also run the latter specification using a village-level quality index,
which is allocated to every household living in the same village. The coefficient is robust to this change.
17
Table 6: First stage regression on grid connection and quality index
instrument: cable density
Dependent variable:
Grid connection
(1)
Cable density
(2)
(3)
Quality of Supply
(4)
(5)
(6)
0.019∗∗∗ 0.018∗∗∗ 0.018∗∗∗ 0.019∗∗∗ 0.019∗∗∗
(0.007)
(0.006)
(0.006)
(0.007)
(0.006)
Size
Children
Hindu
(7)
(8)
(9)
0.021∗∗ 0.020∗∗ 0.020∗∗ 0.021∗∗
(0.008)
(0.008)
(0.008)
(0.008)
(10)
0.020∗∗
(0.008)
0.021∗∗∗ 0.020∗∗∗
(0.001)
(0.001)
0.019∗∗∗
(0.001)
0.007∗∗∗ 0.006∗∗∗
(0.001)
(0.001)
0.007∗∗∗
(0.001)
−0.158∗∗∗−0.154∗∗∗
(0.017)
(0.017)
−0.139∗∗∗
(0.017)
−0.098∗∗∗−0.094∗∗∗
(0.014)
(0.014)
−0.093∗∗∗
(0.014)
0.024
(0.018)
0.023
(0.018)
0.024
(0.018)
0.039
(0.027)
0.005
(0.015)
0.020
(0.028)
Livestock
0.040∗∗∗ 0.068∗∗∗ 0.046∗∗∗
(0.010)
(0.010)
(0.010)
0.037∗∗∗ 0.048∗∗∗
(0.009)
(0.009)
Agriculture
−0.112∗∗∗−0.093∗∗∗
(0.010)
(0.010)
−0.028∗∗∗
(0.008)
Low Skill
−0.058∗∗∗−0.031∗∗∗
(0.010)
(0.010)
−0.010
(0.007)
0.002
(0.008)
0.035
(0.022)
0.042∗
(0.022)
High Skill
0.102
(0.029)
0.013
(0.018)
0.004
(0.015)
Home
∗∗∗
0.023
(0.027)
0.004
(0.015)
∗∗∗
0.116
(0.029)
0.022
(0.018)
0.015
(0.018)
0.040∗∗∗
(0.009)
−0.020∗∗
(0.008)
Year F.E.
Village F.E.
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
Observations
19,570
19,570
19,570
19,489
19,489
14,362
14,362
14,362
14,310
14,310
Notes: All estimations contain a constant. Standard errors in parentheses are clustered at the village level. *** p<0.01,
** p<0.05, * p<0.1. Instrument is the district deviation from the national density of transmission cables per squared
kilometer. Hindu, Home and Livestock are dummy variables. Agriculture, Low Skill and High Skill are occupations
estimated with respect to a benchmark corresponding to any kind of home-based activity, including students and
unemployed.
receiving electricity is associated with 18 percent higher household income. The estimate
is generally robust to the inclusion of household controls for demographic characteristics,
asset ownership and occupations, and various combinations of these controls. Instrumentation of electricity with cable density increases the impact several fold to 180 percent,
implying that receiving electricity almost triples the non-agricultural household income,
and the estimate is largely robust to the inclusion of various household controls.
Returns to electrification show a large increase after instrumentation. A potential
explanation relates to the type of infrastructure reflected in the coefficients. The IV
estimates reflect the impact of the transmission network, while the OLS estimates reflect
both distribution and transmission networks. The smaller OLS estimates may thus be
associated with the efforts of state governments to provide power to poorer areas. The
transmission network, on the other hand, substantially improves the household’s probability of receiving electricity, but it does not reflect the potential downward bias arising
from state policies.
The coefficients of other control variables have the expected signs. Per-adult equivalent household income is negatively associated with household size, with each additional
18
individuals reducing income by 3 to 6 percent. The demographic controls are insignificant,
which is most likely due to the demographic correction in the dependent variable.The
coefficient on livestock ownership is negative, indicating that the household is less likely
to receive income from non-agricultural activities if it owns livestock. Home ownership
has insignificant impact as almost the entire sample are home owners, and the real estate rental market is likely to be very small or nonexistent in these small villages. The
household income is about 41 percent and 125 percent higher than the control group if
the household head was in low-skill and high-skill non-agricultural occupations, respectively. Agriculture is associated with a 12 percent lower non-agricultural income. The
low magnitude of this estimate indicates that households tend to have both agricultural
and non-agricultural incomes and do not necessarily specialize in one type of activity.
These controls for household characteristics are expected to capture differences among
households that may determine whether or not they receive electricity. However, unobserved household characteristics may still determine both income and electrification,
leading to biased estimates. In Columns (11) and (12), we control for household fixed effects to account for these unobserved characteristics. The coefficient on electrification in
the OLS specification has declined to 0.06 percent, while the IV specification is generally
robust at 1.83, both statistically significant at the 5 percent level.
The above exercise is repeated using the variation in land elevation as an instrument
for electrification. The results are presented in Table 8. The magnitudes for the impact
of electrification are quantitatively similar to the previous results. The OLS specification
varies between 0.18 and 0.23, while the IV specification varies between 1.4 and 1.5. The
discrepancy may be due to the fact that 8 coastal districts are dropped from the analysis
as the variation in land elevation could not be measured for these districts.
The model is also estimated by excluding households that live in districts that are
relatively urbanized. The incomes in these districts may be growing faster than others
for reasons that are not related to electricity, but they may be receiving electricity due to
their proximity to cities. This is done by excluding the ten largest cities, and the districts
in which they are located, as well as their neighbouring districts. The 10 biggest cities
in India in decreasing order of importance, are: Mumbai, Delhi, Bangalore, Hyderabad,
Ahmedabad, Chennai, Kolkata, Surat, Pune, and Jaipur.15 The results suggest that the
coefficients of interest, on the Grid connection have roughly the same magnitude and the
same statistical significance as in the whole sample estimation.16
15
This results in elimination of the following districts from the sample: Maharashtra: the districts
of Dhule, Pune, Ahmednagar, Solapur and Satara; Haryana: the districts of Sonipat, Gurgaon and
Faridabad; Karnataka: the district of Bangalore rural, Tumkuru and Kolar; Tamil Nadu: the districts
of Dharmapuri and Kancheepuram; Andhra Pradesh: the district of Medak; Gujarat: the districts of
Ahmedabad, Kheda, Gandhinagar and Bharuch; Rajasthan: the districts of Sikar, Alwar, and Sawai
Madhopur; and finally West Bengal, the districts of South 24 Parganas and North 24 Parganas.
16
These results are presented in Table A.5. The same model is estimated using the land gradient
19
20
19,570
yes
yes
no
8.23
8.58
19,570
yes
yes
no
0.180∗∗∗ 1.807∗∗
(0.023)
(0.793)
(2)
(1)
−0.082
(0.064)
−0.049
(0.044)
19,570
yes
yes
no
8.17
8.36
19,570
yes
yes
no
−0.016
(0.045)
0.023
(0.033)
−0.021
(0.045)
19,570
yes
yes
no
8.06
8.24
19,570
yes
yes
no
−0.156∗∗∗−0.224∗∗∗
(0.019)
(0.041)
−0.417∗∗∗−0.156
(0.036)
(0.142)
−0.403∗∗∗−0.153
(0.036)
(0.143)
0.017
(0.033)
−0.033∗∗∗−0.067∗∗∗
(0.003)
(0.018)
(6)
IVC
−0.037∗∗∗−0.070∗∗∗
(0.003)
(0.018)
(5)
OLS
0.213∗∗∗ 1.900∗∗
(0.023)
(0.868)
(4)
IVC
0.207∗∗∗ 1.788∗∗
(0.023)
(0.855)
(3)
OLS
(8)
IVC
19,489
8.71
9.26
19,489
19,489
yes
yes
no
8.72
9.09
19,489
yes
yes
no
1.246∗∗∗ 1.057∗∗∗
(0.069)
(0.120)
1.259∗∗∗ 1.084∗∗∗
(0.072)
(0.112)
yes
yes
no
0.428∗∗∗ 0.479∗∗∗
(0.023)
(0.036)
0.414∗∗∗ 0.513∗∗∗
(0.023)
(0.052)
yes
yes
no
−0.101∗∗∗ 0.051
(0.024)
(0.081)
−0.036
(0.064)
−0.013
(0.043)
−0.115∗∗∗ 0.077
(0.024)
(0.092)
0.001
(0.043)
0.026
(0.029)
−0.501∗∗∗−0.271∗∗
(0.034)
(0.125)
−0.026∗∗∗−0.057∗∗∗
(0.004)
(0.017)
−0.107∗∗∗−0.182∗∗∗
(0.019)
(0.043)
−0.065
(0.070)
(10)
IVC
0.178∗∗∗ 1.819∗∗
(0.022)
(0.831)
(9)
OLS
−0.139∗∗∗−0.258∗∗∗
(0.019)
(0.058)
0.001
(0.044)
0.165∗∗∗ 1.890∗∗
(0.022)
(0.796)
(7)
OLS
1.872∗∗
(0.881)
(12)
IVC
−0.057
(0.039)
−0.089∗∗
(0.043)
0.123
(0.088)
0.043
(0.092)
19,489
yes
no
yes
7.73
8.19
19,489
yes
no
yes
0.769∗∗∗ 0.830∗∗∗
(0.082)
(0.104)
0.320∗∗∗ 0.294∗∗∗
(0.030)
(0.039)
−0.052∗
(0.030)
−0.034
(0.026)
0.081
(0.054)
0.107
(0.070)
−0.422∗∗∗ −0.214∗
(0.043)
(0.115)
−0.026∗∗∗ −0.044∗∗∗
(0.004)
(0.011)
0.061∗∗
(0.029)
(11)
OLS
Notes: All estimations contain a constant. Standard errors in parentheses are clustered at the village level. *** p<0.01, ** p<0.05, * p<0.1. Income represents total
household income excluding agriculture expressed in per adult equivalent terms. Grid connection is instrumented using the district deviation from the national density of
transmission cables per squared kilometer. Hindu, Home and Livestock are dummy variables. Agriculture, Low Skill and High Skill are occupations estimated with respect
to a benchmark corresponding to any kind of home based activities, including students and unemployed.
F-stat first stage
K-P F-stat (weak ident)
Observations
Year F.E.
Village F.E.
Household F.E.
High Skill
Low Skill
Agriculture
Livestock
Home
Hindu
Children
Size
Grid connection
IVC
OLS
Dependent variable:
Log of income per adult equivalent
Table 7: Effect of a grid connection on income
instrument: density of transmission cables
21
17,282
Observations
17,282
282.34
0.000
17,282
yes
yes
321.72
0.000
17,282
yes
yes
0.020
(0.035)
−0.038
(0.047)
−0.038
(0.048)
17,282
yes
yes
322.27
0.000
17,282
yes
yes
−0.147∗∗∗−0.140∗∗∗
(0.020)
(0.019)
0.026
(0.035)
0.028
(0.036)
(8)
IVL
17,207
17,207
293.38
0.000
17,207
yes
yes
1.227∗∗∗
(0.072)
1.241∗∗∗ 1.212∗∗∗
(0.075)
(0.071)
yes
yes
0.424∗∗∗
(0.024)
0.394∗∗∗ 0.369∗∗∗
(0.025)
(0.024)
yes
yes
−0.111∗∗∗
(0.025)
−0.136∗∗∗−0.158∗∗∗
(0.026)
(0.025)
0.018
(0.047)
−0.096∗∗∗
(0.019)
0.020
(0.048)
0.030
(0.033)
−2.095
(1.419)
−0.023∗∗∗
(0.004)
0.189∗∗∗
(0.023)
(9)
OLS
−0.134∗∗∗−0.131∗∗∗
(0.019)
(0.019)
0.024
(0.049)
0.176∗∗∗ 1.400∗∗∗
(0.023)
(0.061)
(7)
OLS
317.55
0.000
17,207
yes
yes
1.192∗∗∗
(0.068)
0.395∗∗∗
(0.023)
−0.134∗∗∗
(0.024)
−0.092∗∗∗
(0.018)
0.016
(0.047)
0.030
(0.032)
−2.120∗
(1.261)
−0.024∗∗∗
(0.004)
1.411∗∗∗
(0.057)
(10)
IVL
Notes: All estimations contain a constant. Standard errors in parentheses are clustered at the village level. *** p<0.01, ** p<0.05, * p<0.1.
Income represents total household income excluding agriculture expressed in per adult equivalent terms. Grid connection is instrumented using
the district-level variation (standard deviation) in land elevation. The first stage is modelled in a non-linear way using a probit estimation.
Hindu, Home and Livestock are dummy variables. Agriculture, Low Skill and High Skill are occupations estimated with respect to a benchmark
corresponding to any kind of home based activities, including students and unemployed.
Wald chi2 first stage
Wald p-value
yes
yes
Year F.E.
Village F.E.
High Skill
Low Skill
Agriculture
Livestock
Home
yes
yes
0.021
(0.036)
Hindu
−1.711
(1.435)
−1.622
(1.580)
−1.641
(1.445)
−1.547
(1.586)
Size
Grid connection
Children
(6)
IVL
−0.031∗∗∗−0.032∗∗∗
(0.004)
(0.004)
(5)
OLS
−0.034∗∗∗−0.035∗∗∗
(0.004)
(0.004)
(4)
IVL
0.225∗∗∗ 1.493∗∗∗
(0.024)
(0.060)
(3)
OLS
0.219∗∗∗ 1.491∗∗∗
(0.024)
(0.060)
(2)
(1)
0.192∗∗∗ 1.457∗∗∗
(0.024)
(0.065)
IVL
OLS
Dependent variable:
Log of income per adult equivalent
Table 8: Effect of a grid connection on income
instrument: district level variation in land elevation
Given that electrification rates increased by 16 percentage points between 1994 and
2005, these results imply that non-agricultural household incomes increased by 22 to 29
percent over this 11-year period as a result of increased access to electricity ((1.4)(0.16) to
(1.89)(0.16)). A household moving from no electricity to having grid access, experiences
an annual increase in income of 2.5%. However, this result only shows the impact of
moving from no electricity to some connection, with no assumption on the quality of
electricity supply.
4.2
Intensive Margin: Quality of Electricity Supply
Table 9 shows the impact of mean quality of electricity supply on income. The results
for OLS and IV estimations are reported, with transmission cable density as instrument.
In an alternative specification, both the land elevation and transmission cable density
were used as instruments. In all specifications standard errors are robust and clustered
at the village level, and all specifications contain village and year fixed effects. Columns
(1) to (3) include only household composition controls, and columns (4) to (6) include all
household-level controls. The IV estimation results show cable density as the single IV
(in columns 2 and 5) and results for both IVs are reported in columns 3 and 6. Columns
(7) and (8) report results for OLS and single IV using household fixed effects instead of
village fixed effects.
The coefficient on column (5), implies that the income of an household whose quality
of supply increases from 0 to 1 goes up by a factor of 2.6. The results suggest that
OLS significantly underestimates the impact of electrification, implying that households
who received an improved quality of electricity supply between 1994 and 2005 are living
in under-performing areas with slower growth and economic improvement rates. This
phenomenon is observed also in Dinkelman (2012) and in Lipscomb et al. (2013).
From Table 2, we know that the quality of supply improved by 32 percentage points on
average across the country between 1994 and 2005. These numbers imply that household
non-agricultural incomes increased by roughly 85% percent over this 11-year period as a
result of increased quality of power supply (2.649)(0.32). This suggests that the annual
improvement attributable to improvements in quality of power supply is roughly 7.8%.
The data on the quality of power supply has some missing values at the household
level, which can lead to biased estimates if non-reporting households have common characteristics correlated with income or electrification. To check if this is affecting our
results, the same model is also estimated using an aggregated quality measure at the village level and attributing the average quality to all households living in the village. The
magnitude and the significance of coefficients are almost identical to the main results,
instrument, but not presented for brevity.
22
implying that the missing data does not pose a problem in identifying the coefficient.17
Table 9: Effect of quality of electricity supply at the household level on household
income
Dependent variable:
Log of income per adult equivalent
Quality
OLS
IVC
IVC+L
OLS
IVC
IVC+L
OLS
IVC
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
0.185∗∗∗ 2.733∗
(0.037)
(1.471)
2.751∗
(1.429)
0.163∗∗∗ 2.649∗
(0.036)
(1.437)
2.615∗
(1.376)
0.130∗∗∗
(0.047)
2.642∗
(1.423)
Size
−0.035∗∗∗−0.055∗∗∗ −0.055∗∗∗
(0.004)
(0.013)
(0.013)
−0.023∗∗∗−0.041∗∗∗ −0.041∗∗∗
(0.004)
(0.012)
(0.012)
−0.023∗∗∗
(0.005)
−0.035∗∗∗
(0.011)
Children
−0.461∗∗∗−0.197
(0.042)
(0.164)
−0.561∗∗∗−0.316∗∗ −0.319∗∗
(0.041)
(0.154)
(0.148)
−0.440∗∗∗
(0.052)
−0.207
(0.152)
Hindu
0.064
(0.048)
0.037
(0.055)
−0.196
(0.160)
0.037
(0.055)
Home
0.058
(0.042)
0.037
(0.049)
0.037
(0.048)
0.200∗
(0.103)
0.300∗∗
(0.135)
0.054
(0.058)
0.002
(0.074)
0.003
(0.073)
0.167∗∗
(0.071)
0.070
(0.115)
Livestock
−0.115∗∗∗−0.213∗∗∗ −0.212∗∗∗
(0.022)
(0.063)
(0.061)
−0.037
(0.032)
−0.136∗
(0.070)
Agriculture
−0.140∗∗∗−0.087∗
(0.029)
(0.048)
−0.086∗∗
(0.036)
−0.145∗∗
(0.058)
−0.088∗
(0.047)
Low Skill
0.421∗∗∗ 0.414∗∗∗ 0.414∗∗∗
(0.028)
(0.033)
(0.033)
0.298∗∗∗
(0.035)
0.207∗∗∗
(0.067)
High Skill
1.263∗∗∗ 1.142∗∗∗ 1.143∗∗∗
(0.074)
(0.116)
(0.114)
0.816∗∗∗
(0.090)
0.861∗∗∗
(0.131)
Year F.E.
Village F.E.
Household F.E.
Observations
F-stat first stage
K-P F-stat (weak ident)
yes
yes
no
yes
yes
no
yes
yes
no
yes
yes
no
yes
yes
no
yes
yes
no
yes
no
yes
yes
no
yes
12,652
12,652
12,652
12,605
12,605
12,605
12,605
12,605
4.90
5.19
5.71
2.92
5.07
5.4
5.96
3.07
5.02
5.61
Notes: All estimations contain a constant. Standard errors in parentheses are clustered at the village level. ***
p<0.01, ** p<0.05, * p<0.1. Income represents total household income excluding agriculture expressed in per adult
equivalent terms. Instruments used are density of transmission cables (IVC ) and both the cable density and variation
in land elevation (IVC+L ). Hindu, Home and Livestock are dummy variables. Agriculture, Low Skill and High Skill
are occupations estimated with respect to a benchmark corresponding to any kind of home based activities, including
students and unemployed. The Quality takes value 0 for no electricity, 1 for high quality electricity supply and 0.5
for low quality electricity supply. The threshold for high quality power supply in 2005 is fixed at an average of 18
hours of power per day.
In order to better understand the functional relationship between power quality and
household income, we run a kernel-weighted local regression of household income on the
predicted average quality of electricity received by the household obtained from the first
stage. The results, presented in Figure 5, show that there is a linear and monotonic
relationship, with high income households receiving higher quality of electricity, where
the quality is measured by the number of hours of electricity received per day. Starting
from no grid connection, the first five hours of electricity exhibit high marginal returns.
Even if electrification is provided only for a few hours a day, the initial access to electricity
is associated with higher incomes. Between 5 and 20 hours per day, incomes roughly stay
17
The results are presented in Table A.2
23
constant, but increase again beyond 20 hours of supply. That is, at very high levels of
quality, marginal returns are high.
Figure 5: Household income as a function of daily electricity supply, 2005
9.5
10
ln(income)
Log of income 10.5
11
ln(income)
11
10.5
9.5
0
1
5
10
15
20
25
Predicted
kernel
0
= epanechnikov,
hours
degree = 0, bandwidth = 1.27, pwidth = 1.91
0
5
10
15
Predicted hours
kernel = epanechnikov, degree = 0, bandwidth
= 1.27,
pwidth = 1.91
Predicted hours 20
25
Note: Kernel weighted polynomial regression with 95% confidence band.
This implies that the impact of electrification on households cannot be considered
independently of the intensive margin. While the initial connection is important in the
sense that it induces reallocation of labor and capital within the household, the same
magnitude of impact can be observed at the other end of the scale. Starting from a grid
connection, the impact of moving from low quality to high quality electricity can be as
large as the initial connection. This highlights the importance of providing a high quality
supply of power, as the potential benefits of electricity are not completely realized by
just providing a grid connection and low quality power.
The slope of the smoothed curve seems to change at about 20 hours per day. Recall
our choice of 18 hours a day as the threshold for high quality. In order to assess whether
or not these results are robust to this choice, the main model is estimated with 16 and
20 hours as thresholds for high quality power supply. Reducing the threshold to 16
hours per day does not affect the results. However, increasing the threshold to 20 hours
24
increases the magnitudes of the coefficient to 3.4, indicating that the impact of quality
is not constant across the quality spectrum.
The quality of power supply may be higher in relatively urbanized areas due to a
higher number of power plants or higher investments in power transmission. We reestimate the model by excluding the districts with ten largest cities as well as their
neighboring districts, as in section 4.1. The results suggest that the coefficients of interest,
on Quality generally have the same magnitude and the same statistical significance as in
the whole sample estimation.18
4.3
District Level Analysis
The above analysis identifies the impact of electrification at the household level controlling for village effects and household characteristics. This specification can potentially
yield biased results if there are unobserved factors that drive both electrification outcomes
and household income. For this reason, the model is also estimated by aggregating the
outcomes to the district level using sampling weights.19 The variables for electrification
outcomes are (i) grid connection rate and (ii) average quality of power within the district.
This model is estimated using district fixed effects to account for time-invariant factors
across districts, and year fixed effects to account for changes that are common to all
districts.
The results are presented in Table 10. The OLS specification yields insignificant results although the magnitude is similar to those reported earlier. The instrumentation
of electrification with density of transmission cables yields significant and higher coefficients. Overall, the results are not altered by aggregating the data to the district level.
Higher electrification rates and higher quality of power supply within districts increases
the households’ non-agricultural income.
It is possible that districts in relatively urbanized and industrial areas receive better
electricity, and exhibit higher income growth due to their proximity to large cities where
most economic activity takes place. As before, the model is also estimated by excluding
urban centers, specifically districts containing the 10 largest cities and their neighboring
districts. The results presented in Table 11 show that the results are robust to this
change. This is likely due to the fact that the villages in our sample are located in
remote areas with limited access to urban centers.
18
19
These results are presented in Table A.6
This can not be done at the village level as the location of villages is unknown.
25
Table 10: District level effects on income
instrument: density of transmission cables
Dependent variable:
Log of district-level mean household income
Grid connection
Share of district electrification
Quality
OLS
IVC
OLS
IVc
(1)
(2)
(3)
(4)
0.225
(0.193)
1.567∗∗
(0.613)
District-level quality
0.147
(0.178)
2.219∗
(1.331)
Year F.E.
District F.E.
yes
yes
yes
yes
yes
yes
yes
yes
Observations
314
314
314
314
F-stat first stage
K-P F-stat (weak ident)
8.443
6.632
5.416
3.547
Notes:All estimations contain a constant. Standard errors in parentheses are
robust. *** p<0.01, ** p<0.05, * p<0.1. Income represents the district-level
average of total household income excluding agriculture expressed in per adult
equivalent terms. The Share of district electrification and the District-level
quality are instrumented using the density of cables.
Table 11: District level analysis (excluding urban centers)
instrument: density of transmission cables
Dependent variable:
Log of district-level mean household income
Grid connection
Share of district electrification
Quality
OLS
IVC
OLS
IVc
(1)
(2)
(3)
(4)
0.276
(0.203)
1.868∗∗∗
(0.718)
District-level quality
0.263
(0.190)
Year F.E.
District F.E.
yes
yes
Observations
276
F-stat first stage
K-P F-stat (weak ident)
yes
yes
yes
yes
276
276
6.573
4.982
2.341∗
(1.401)
yes
yes
276
4.968
3.206
Notes:All estimations contain a constant. Standard errors in parentheses are
robust. *** p<0.01, ** p<0.05, * p<0.1. Income represents the district-level
average of total household income excluding agriculture expressed in per adult
equivalent terms. The Share of district electrification and the District-level
quality are instrumented using the density of cables.
26
5
Potential Channels
How does electrification increase non-agricultural income? While data restrictions do not
allow for a detailed analysis of potential channels, it is possible to provide some insight by
decomposing income into two components: (i) wage income that includes the formal and
informal sectors, and (ii) business income, which includes family business, petty trades
and other home-based activities. Increased productivity in agricultural production due
to better access to electricity is likely to increase labor demand and wages. This can
yield differential effects in wages if the labor mobility across sectors is imperfect, which
may be the case if individuals in agriculture have little or no training in business, for
example.
An important issue in studying these channels is that the selection into wage or
business activities is not likely to be random. It can be driven by factors such as land
ownership, demographic structure within the household, or current occupation. In addition, the data consists of many households who do not report both types of incomes.
The model is therefore estimated using a Tobit specification, which implies the existence
of a latent variable Yit∗ , so that the income we observe, Yit , is given by
(
Yit =
Yit∗
0
if Yit∗ > 0
if Yit∗ ≤ 0
(3)
We use the transmission cable density and variation in land elevation as instruments
for electrification.20 The results are presented in Table 12. The estimated coefficient
represents the effect of electrification on the latent variable, and so it incorporates the
mechanisms that cause households to move in and out of these activities. These results
suggest that receiving electricity leads to an increase in wage incomes but also a significant
reduction in business incomes.
This duality can be explained by a model with immobile labor, and perfect mobility of
other factors of production. A disproportional increase in agricultural productivity would
move factors of production into agriculture, increasing wages but reducing the returns to
business activities. Alternatively, individuals may be allocating more labor hours toward
wage activities, and not in the ownership of businesses. In this case, mobility of labor
would equalize wages across activities, but would have differential effects on earnings
through changes in labor supply. In both scenarios, the earnings through wages will
increase but earnings through business income may decrease.
Similar results hold for the quality of power supply. The higher quality of power
supply within village increases wage income, but again reduces business income. Again,
20
OLS results and the estimation with only the variation in land elevation instrument yields similar
results, but are not presented here.
27
the mechanisms driving this result may work through wages or labor supply, which we
are unable to identify due to lack of better data.
Table 12: Effect of grid connection and quality of electricity supply on wages and business
income
Dependent variable:
Log of income per adult equivalent
Effect of grid connection
Wage and salary
income
Grid connection
Effect of quality of supply
Business
income
Wage and salary
income
Business
income
IVC
IVC+L
IVC
IVC+L
IVC
IVC+L
IVC
IVC+L
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
−8.342∗
(4.630)
−6.094
(4.075)
1.539∗∗∗ 1.578∗∗∗
(0.573)
(0.536)
1.697∗∗∗ 1.697∗∗∗
(0.497)
(0.497)
Village level
Quality
−9.195∗∗
(4.634)
−9.195∗∗
(4.634)
−0.013
(0.011)
−0.014
(0.011)
0.248∗∗∗ 0.205∗∗
(0.094)
(0.083)
0.016∗∗∗ 0.016∗∗∗
(0.002)
(0.002)
0.090∗∗∗
(0.021)
0.090∗∗∗
(0.021)
Children
0.010
(0.089)
0.016
(0.085)
−2.089∗∗∗ −1.773∗∗∗
(0.732)
(0.654)
−0.194∗∗∗ −0.194∗∗∗
(0.031)
(0.031)
−0.976∗∗∗
(0.291)
−0.976∗∗∗
(0.291)
Hindu
0.019
(0.036)
0.018
(0.036)
−0.438
(0.282)
−0.491∗
(0.267)
−0.639∗∗∗
(0.238)
−0.639∗∗∗
(0.238)
Home
−0.089
(0.054)
−0.090∗
(0.055)
0.887∗
(0.455)
0.819∗
(0.432)
Livestock
−0.211∗∗∗ −0.213∗∗∗
(0.033)
(0.032)
0.757∗∗∗ 0.657∗∗∗
(0.271)
(0.247)
Agriculture
1.008∗∗∗ 1.011∗∗∗
(0.057)
(0.054)
Low Skill
High Skill
Size
0.058∗∗
(0.026)
−0.032
(0.040)
0.058∗∗
(0.026)
−0.032
(0.040)
0.606
(0.394)
0.606
(0.394)
−0.168∗∗∗ −0.168∗∗∗
(0.018)
(0.018)
0.517∗∗∗
(0.172)
0.517∗∗∗
(0.172)
−7.047∗∗∗ −6.842∗∗∗
(0.680)
(0.627)
0.868∗∗∗ 0.868∗∗∗
(0.019)
(0.019)
−6.304∗∗∗
(0.455)
−6.304∗∗∗
(0.455)
1.085∗∗∗ 1.086∗∗∗
(0.030)
(0.030)
−4.253∗∗∗ −4.185∗∗∗
(0.375)
(0.354)
1.034∗∗∗ 1.034∗∗∗
(0.020)
(0.020)
−3.990∗∗∗
(0.308)
−3.990∗∗∗
(0.308)
0.954∗∗∗ 0.949∗∗∗
(0.099)
(0.097)
−1.131
(0.774)
1.123∗∗∗ 1.123∗∗∗
(0.056)
(0.056)
−2.059∗∗∗
(0.503)
−2.059∗∗∗
(0.503)
−1.399∗∗
(0.712)
Year F.E.
Village F.E.
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
Observations
17,217
17,217
17,217
17,217
17,217
17,217
17,217
17,217
Wald chi2 first stage
Wald p-value
15.61
0.000
19.03
0.000
5.36
0.021
3.65
0.056
13.47
0.000
13.47
0.000
4.71
0.029
4.71
0.029
Notes: All estimations are tobit IV specifications, performed using the two steps methodology. All specifications contain
a constant. Standard errors in parentheses are robust. *** p<0.01, ** p<0.05, * p<0.1. Income corresponds to the total
household income coming respectively from wage and salary activities or business activities and is computed in per adult
equivalent terms. Electrification and the quality are instrumented using the density of transmission cables (IVC ) and a
combination of the normalized density of transmission cables and the district level variation in land elevation (IVC+L ).
Hindu, Home and Livestock are dummy variables. Agriculture, Low Skill and High Skill are occupations estimated with
respect to a benchmark corresponding to any kind of home based activities, including students and unemployed. The Village
level quality takes value 0 for no electricity, 1 for high quality electricity supply and 0.5 for low quality electricity supply.
6
Concluding Remarks
Previous studies of the effect of electrification have focused on the extensive margin. In
this paper we examine the effect of a grid connection as well as the quality of power
supply on household income in India. We use a nationally representative dataset to show
that the quality of electricity supply may be as important as connecting to the grid. A
28
grid connection increases non-agricultural incomes of rural households by about 20-30
percent during the study period (1994-2005). However, higher quality of electricity (in
terms of fewer outages and more hours per day) increased non-ag incomes by about 85
percent in the study period. This is equivalent to an income growth of about 8 percent
per year.
The result show that returns from electricity are high for the first few hours of supply,
and when the supply is close to the maximum 24 hours. On the other hand, when average
power supply is within a large middle band (5-20 hours a day), returns to household
income are relatively flat. From a policy point of view, this suggests that while a grid
connection brings high benefits at low levels of quality, further increases in supply bring
no increased benefits unless the supply is of very high quality.
Our results suggest that policies that aim to provide reliable electricity to households
may bring about significant economic benefits. These estimates can be used to perform
benefit-cost analysis for infrastructure improvements both at the extensive and intensive
margins.
References
[1] The OECD List of Social Indicators. OECD, Paris, 1982.
[2] World Energy Assessment: Energy and the Challenge of Sustainability. United Nations
Development Program (UNDP), New York, 2000.
[3] World Energy Outlook 2011. International Energy Agency, 2011.
[4] J.D. Angrist. Lifetime Earnings and the Vietnam Era Draft Lottery: Evidence from Social
Security Administrative Records. American Economic Review, 80(3):313–36, June 1990.
[5] J.D. Angrist. Estimating the Labor Market Impact of Voluntary Military Service Using
Social Security Data on Military Applicants. Econometrica, 66(2):249–288, March 1998.
[6] D.A. Aschauer. Is Public Expenditure Productive?
23(2):177–200, March 1989.
Journal of Monetary Economics,
[7] P. Balachandra. Dynamics of Rural Energy Access in India: An Assessment. Energy,
36(9):5556 – 5567, 2011.
[8] G. Bensch, J. Kluve, and J. Peters. Rural Electrification in Rwanda - An Impact Assessment
Using Matching Techniques. Ruhr Ecnomic Papers 231, RUHR, December 2010.
[9] A. Bhide and C. Rodriguez Monroy. Energy Poverty: A Special Focus on Energy Poverty
in India and Renewable Energy Technologies. Renewable and Sustainable Energy Reviews,
15(2):1057 – 1066, 2011.
29
[10] M.H. Brown and R.P. Sedano. Electricity Transmission, A Primer. National Council on
Electricity Policy, June 2004.
[11] Taryn Dinkelman. The Effects of Rural Electrification on Employment: New Evidence from
South Africa. American Economic Review, 101(7):3078–3108, December 2011.
[12] E. Duflo and R. Pande. Dams. The Quarterly Journal of Economics, 122(2):601–646, 05
2007.
[13] T. Garcia-Mila and T.J. McGuire. The Contribution of Publicly Provided Inputs to States’
Economies. Regional Science and Urban Economics, 22(2):229–41, June 1992.
[14] D. Holtz-Eakin. State-Specific Estimates of State and Local Government Capital. Regional
Science and Urban Economics, 23(2):185–209, April 1993.
[15] G.W. Imbens and J.D. Angrist. Identification and Estimation of Local Average Treatment
Effects. Econometrica, 62(2):467–75, March 1994.
[16] S.R. Khandker, D.F. Barnes, and H.A. Samad. Welfare Implications of Rural Electrification,
A Case Study for Bangladesh. Policy Research Working Paper 4859, World Bank, March
2009a.
[17] S.R. Khandker, D.F. Barnes, and H.A. Samad. Energy Poverty in Rural and Urban India.
are the Energy Poor also Income Poor? Policy Research Working Paper 5463, World Bank,
November 2010.
[18] S.R. Khandker, D.F. Barnes, H.A. Samad, and N.H. Minh. Welfare Implications of Rural
Electrification, Evidence from Vietnam. Policy Research Working Paper 5057, World Bank,
September 2009b.
[19] M. Lipscomb, A.M. Mobarak, and T. Barham. Development Effects of Electrification:
Evidence from the Geologic Placement of Hydropower Plants in Brazil. American Economic
Journal: Applied Economics, April 2013.
[20] V. Modi. Improving Electricity Services in Rural India. Working Paper 30, Center on
Globalization and Sustainable Development, The Earth Institute at Columbia University,
December 2005.
[21] H. Oda and Y. Tsujita. The Determinants of Rural Electrification: The Case of Bihar,
India. Energy Policy, 39(6):3086–3095, June 2011.
[22] Directorate of Census Office of the Registrar General, editor. Census of India 2001. Government of India, New Delhi, India, 2001.
[23] L.-H. Roller and L. Waverman. Telecommunications Infrastructure and Economic Development: A Simultaneous Approach. American Economic Review, 91(4):909–923, September
2001.
30
[24] J.P. Rud. Electricity Provision and Industrial Development: Evidence from India. Journal
of Development Economics, 97(2):352 – 367, 2012.
[25] USGS. Global Land Cover Facility (SRTM). In Shuttle Radar Topography Mission. Global
Land Cover Facility, University of Maryland, 2004.
[26] H. Yang. Overview of the Chinese Electricity Industry and its Current Uses. Working Paper
0617, CWPE, February 2006.
31
A
Appendix Tables
Table A.1: Occupational Categories
#
1
New Group
agriculture
2
low skill
1994 Group
cultivation, allied agricultural activities, agricultural wage labor, cattle
tending
artisan/independent
work,
petty
shop/other small business, organized
business/trade,
domestic servant,
salaried employment/pension, nonagricultural wage labor,
3
high skill
qualified profession/not classified
4
other
own household work, student, family
work, child (<15), living on income
from rent, int, div, unemployed, any
other, not willing to work, not specified
32
2005 Group
farm manager, cultivator, other farmers, ag labor, plantation lab, other farm
money lenders, house keepers, maids,
sweepers, launderers, elected officials,
govt officials, clerical supe, village officials, clerical nec, typist, book-keepers,
computing op, transp/commun supe,
transp conductors,
mail distributor, telephone op, shopkeepers,
manuf agents, technical sales, sales,
shop, FIRE sales, salesa, nec, hotel/restaurant, cooks/waiters, police,
service nec, barbers, tanners, tailors,
electrical, plumbers/welders, jewellery,
shoe makers, carpenters, stone cutters, construction, forestry, hunters,
fishermen, miners, metal workers,
wood/paper, chemical, textile, foods,
tobacco, machine tool op, assemblers,
cinema op, potters, rubber/plastic,
paper, printing, painters, production
nec, boilermen, loaders, drivers, labour
nec
physical sci tech, engineers, eng
tech, air/ship officer, life sciantists,
life science tech, physicians, nursing, other scientific, statisticians,
economists, accountants, social scientists, lawyers, teachers, journalists,
mgr whsl/retail, mgr finance, mgr
manf, mgr transp/commun, mgr
service, managerial nec
n/a
Table A.2: Effect of quality of electricity supply – Village Level
Dependent variable:
Log of income per adult equivalent
Village level
quality
OLS
IVC
IVC+L
OLS
IVC
IVC+L
OLS
IVC
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
0.229∗∗∗ 1.965∗
(0.084)
(1.123)
1.966∗
(1.123)
0.232∗∗∗ 2.065∗
(0.084)
(1.121)
2.066∗
(1.121)
0.235∗∗∗
(0.082)
2.062∗
(1.111)
Size
−0.032∗∗∗−0.033∗∗∗ −0.033∗∗∗
(0.004)
(0.004)
(0.004)
−0.022∗∗∗−0.023∗∗∗ −0.023∗∗∗
(0.004)
(0.004)
(0.004)
−0.025∗∗∗
(0.005)
−0.026∗∗∗
(0.005)
Children
−0.461∗∗∗−0.446∗∗∗ −0.446∗∗∗
(0.037)
(0.039)
(0.039)
−0.551∗∗∗−0.536∗∗∗ −0.536∗∗∗
(0.035)
(0.037)
(0.037)
−0.460∗∗∗
(0.045)
−0.436∗∗∗
(0.051)
Hindu
0.032
(0.037)
0.033
(0.036)
0.033
(0.036)
Home
0.040
(0.034)
0.043
(0.033)
0.043
(0.033)
0.171∗∗
(0.076)
0.178∗∗
(0.075)
0.025
(0.048)
0.039
(0.058)
0.039
(0.058)
0.122∗∗
(0.059)
0.149
(0.093)
Livestock
−0.099∗∗∗−0.127∗∗∗ −0.127∗∗∗
(0.019)
(0.026)
(0.026)
−0.044∗
(0.026)
−0.096∗∗
(0.043)
Agriculture
−0.140∗∗∗−0.142∗∗∗ −0.142∗∗∗
(0.026)
(0.029)
(0.029)
−0.073∗∗
(0.032)
−0.073∗
(0.038)
Low Skill
0.404∗∗∗ 0.399∗∗∗ 0.399∗∗∗
(0.025)
(0.025)
(0.025)
0.305∗∗∗
(0.032)
0.297∗∗∗
(0.034)
High Skill
1.247∗∗∗ 1.230∗∗∗ 1.230∗∗∗
(0.073)
(0.075)
(0.075)
0.755∗∗∗
(0.085)
0.726∗∗∗
(0.094)
Year F.E.
Village F.E.
Household F.E.
Observations
F-stat first stage
K-P F-stat (weak ident)
yes
yes
no
yes
yes
no
yes
yes
no
yes
yes
no
yes
yes
no
yes
yes
no
yes
no
yes
yes
no
yes
17,302
17,302
17,302
17,226
17,226
17,226
17,226
17,226
6.53
6.96
7.21
3.88
6.54
6.99
8.63
4.66
6.54
7.25
Notes: All estimations contain a constant. Standard errors in parentheses are clustered at the village level. ***
p<0.01, ** p<0.05, * p<0.1. Income represents total household income excluding agriculture expressed in per
adult equivalent terms. Instruments used are density of transmission cables (IVC ) and both the cable density and
variation in land elevation (IVC+L ). Hindu, Home and Livestock are dummy variables. Agriculture, Low Skill
and High Skill are occupations estimated with respect to a benchmark corresponding to any kind of home based
activities, including students and unemployed. The Village level quality takes value 0 for no electricity, 1 for high
quality electricity supply and 0.5 for low quality electricity supply. The threshold for high quality power supply in
2005 is fixed at an average of 18 hours of power per day.
33
Table A.3: Effect of quality of electricity supply - 16 hour threshold
Dependent variable:
Log of income per adult equivalent
Quality
OLS
IVC
IVC+L
OLS
IVC
IVC+L
(1)
(2)
(3)
(4)
(5)
(6)
0.176∗∗∗ 2.636∗
(0.036)
(1.384)
2.635∗
(1.381)
0.151∗∗∗ 2.538∗
(0.035)
(1.351)
2.535∗
(1.346)
Size
−0.032∗∗∗−0.053∗∗∗ −0.053∗∗∗
(0.004)
(0.013)
(0.013)
−0.020∗∗∗−0.039∗∗∗ −0.038∗∗∗
(0.004)
(0.012)
(0.012)
Children
−1.699
(1.479)
−0.413
(1.317)
−0.414
(1.316)
−2.206∗
(1.318)
−0.994
(1.172)
−0.995
(1.170)
0.055
(0.047)
0.022
(0.055)
0.022
(0.055)
0.047
(0.042)
0.019
(0.049)
0.019
(0.049)
0.058
(0.058)
0.021
(0.070)
0.021
(0.070)
Hindu
Home
Livestock
−0.109∗∗∗−0.197∗∗∗ −0.197∗∗∗
(0.022)
(0.057)
(0.057)
Agriculture
−0.129∗∗∗−0.078∗
(0.029)
(0.047)
−0.078∗
(0.047)
Low Skill
0.436∗∗∗ 0.427∗∗∗ 0.427∗∗∗
(0.028)
(0.033)
(0.033)
High Skill
1.253∗∗∗ 1.115∗∗∗ 1.115∗∗∗
(0.074)
(0.119)
(0.119)
Year F.E.
Village F.E.
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
Observations
12,632
12,632
12,632
12,585
12,585
12,585
5.28
5.75
2.68
2.92
5.53
6.01
2.82
3.07
F-stat first stage
K-P F-stat (weak ident)
Notes: The threshold for high quality of supply in 2005 is fixed at an average of 16
hours of power per day. All estimations contain a constant. Standard errors in parentheses are clustered at the village level. *** p<0.01, ** p<0.05, * p<0.1. Income
corresponds to the total household income excluding income coming from agricultural
activities and is computed in per adult equivalent terms. The Quality is instrumented
using the district deviation from the normalized density of transmission cables (IVC )
and with the density of transmission cables and the district-level variation in land
elevation (IVC+L ). Hindu, Home and Livestock are dummy variables. Agriculture,
Low Skill and High Skill are occupations estimated with respect to a benchmark corresponding to any kind of home based activities, including students and unemployed.
The Quality takes value 0 for no electricity, 1 for high quality electricity supply and
0.5 for low quality electricity supply.
34
Table A.4: Effect of quality of electricity supply - 20 hour threshold
Dependent variable:
Log of income per adult equivalent
Quality
OLS
IVC
IVC+L
OLS
IVC
IVC+L
(1)
(2)
(3)
(4)
(5)
(6)
0.233∗∗∗ 3.478∗
(0.038)
(2.055)
3.379∗
(1.981)
0.201∗∗∗ 3.399∗
(0.036)
(2.051)
3.350∗
(1.988)
Size
−0.032∗∗∗−0.059∗∗∗ −0.058∗∗∗
(0.004)
(0.018)
(0.018)
−0.020∗∗∗−0.045∗∗ −0.045∗∗∗
(0.004)
(0.017)
(0.017)
Children
−1.688
(1.439)
−0.187
(1.091)
−0.233
(1.068)
−2.201∗
(1.287)
−0.721
(1.061)
−0.744
(1.036)
0.056
(0.047)
0.027
(0.059)
0.028
(0.058)
0.049
(0.041)
0.025
(0.055)
0.026
(0.054)
0.054
(0.058)
0.033
(0.076)
0.034
(0.075)
Hindu
Home
Livestock
−0.114∗∗∗−0.244∗∗∗ −0.242∗∗∗
(0.022)
(0.088)
(0.086)
Agriculture
−0.127∗∗∗−0.107∗∗∗ −0.107∗∗∗
(0.029)
(0.040)
(0.040)
Low Skill
0.437∗∗∗ 0.386∗∗∗ 0.387∗∗∗
(0.028)
(0.046)
(0.045)
High Skill
1.251∗∗∗ 0.991∗∗∗ 0.995∗∗∗
(0.074)
(0.200)
(0.196)
Year F.E.
Village F.E.
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
yes
Observations
12,632
12,632
12,632
12,585
12,585
12,585
3.36
3.72
1.78
1.98
3.43
3.80
1.85
2.07
F-stat first stage
K-P F-stat (weak ident)
Notes: The threshold for high quality of supply in 2005 is fixed at an average of 20
hours of power per day. All estimations contain a constant. Standard errors in parentheses are clustered at the village level. *** p<0.01, ** p<0.05, * p<0.1. Income
corresponds to the total household income excluding income coming from agricultural
activities and is computed in per adult equivalent terms. The Quality is instrumented
using the district deviation from the normalized density of transmission cables (IVC )
and with the density of transmission cables and the district-level variation in land
elevation (IVC+L ). Hindu, Home and Livestock are dummy variables. Agriculture,
Low Skill and High Skill are occupations estimated with respect to a benchmark corresponding to any kind of home based activities, including students and unemployed.
The Quality takes value 0 for no electricity, 1 for high quality electricity supply and
0.5 for low quality electricity supply.
35
36
17,038
yes
yes
no
6.62
6.91
17,038
yes
yes
no
0.176∗∗∗ 1.990∗∗
(0.025)
(0.862)
(2)
(1)
−0.054
(0.075)
−0.023
(0.050)
17,038
yes
yes
no
6.76
6.94
17,038
yes
yes
no
−0.031
(0.049)
0.024
(0.032)
−0.034
(0.048)
17,038
yes
yes
no
6.60
6.77
17,038
yes
yes
no
−0.146∗∗∗−0.234∗∗∗
(0.021)
(0.049)
−0.417∗∗∗−0.118
(0.038)
(0.160)
−0.405∗∗∗−0.119
(0.038)
(0.160)
0.019
(0.032)
−0.035∗∗∗−0.072∗∗∗
(0.004)
(0.019)
(6)
IVC
−0.038∗∗∗−0.075∗∗∗
(0.004)
(0.020)
(5)
OLS
0.210∗∗∗ 2.082∗∗
(0.025)
(0.933)
(4)
IVC
0.204∗∗∗ 1.946∗∗
(0.025)
(0.910)
(3)
OLS
(8)
IVC
16,965
7.01
7.45
16,965
16,965
yes
yes
no
7.21
7.53
16,965
yes
yes
no
1.227∗∗∗ 0.979∗∗∗
(0.074)
(0.144)
1.239∗∗∗ 1.003∗∗∗
(0.077)
(0.135)
yes
yes
no
0.429∗∗∗ 0.490∗∗∗
(0.024)
(0.041)
0.413∗∗∗ 0.533∗∗∗
(0.025)
(0.061)
yes
yes
no
−0.090∗∗∗ 0.071
(0.026)
(0.084)
−0.011
(0.075)
−0.031
(0.048)
−0.107∗∗∗ 0.101
(0.027)
(0.098)
0.024
(0.049)
0.025
(0.029)
−0.505∗∗∗−0.239∗
(0.036)
(0.142)
−0.027∗∗∗−0.062∗∗∗
(0.004)
(0.018)
−0.099∗∗∗−0.195∗∗∗
(0.020)
(0.051)
−0.051
(0.083)
(10)
IVC
0.178∗∗∗ 2.016∗∗
(0.024)
(0.904)
(9)
OLS
−0.131∗∗∗−0.276∗∗∗
(0.020)
(0.069)
0.020
(0.050)
0.165∗∗∗ 2.108∗∗
(0.024)
(0.883)
(7)
OLS
2.088∗∗
(0.983)
(12)
IVC
−0.066
(0.043)
16,965
yes
no
yes
6.17
6.55
16,965
yes
no
yes
0.721∗∗∗ 0.766∗∗∗
(0.086)
(0.112)
0.316∗∗∗ 0.292∗∗∗
(0.031)
(0.042)
−0.062∗
(0.032)
−0.107∗
(0.055)
0.125
(0.105)
0.097∗
(0.059)
−0.027
(0.028)
−0.007
(0.109)
0.107
(0.075)
−0.428∗∗∗ −0.195
(0.045)
(0.129)
−0.026∗∗∗ −0.045∗∗∗
(0.005)
(0.012)
0.061∗∗
(0.030)
(11)
OLS
Notes: All estimations contain a constant. Standard errors in parentheses are clustered at the village level. *** p<0.01, ** p<0.05, * p<0.1. Income represents total
household income excluding agriculture expressed in per adult equivalent terms. Electrification is instrumented using the district deviation from the national density of
transmission cables per squared kilometer. Hindu, Home and Livestock are dummy variables. Agriculture, Low Skill and High Skill are occupations estimated with respect
to a benchmark corresponding to any kind of home based activities, including students and unemployed.
F-stat first stage
K-P F-stat (weak ident)
Observations
Year F.E.
Village F.E.
Household F.E.
High Skill
Low Skill
Agriculture
Livestock
Home
Hindu
Children
Size
Grid connection
IVC
OLS
Dependent variable:
Log of income per adult equivalent
Table A.5: Effect of a grid connection on income (excluding urban centers)
instrument: density of transmission cables
Table A.6: Effect of quality of electricity supply at the household-level on household
income (excluding urban centers)
Dependent variable:
Log of income per adult equivalent
Quality
Size
Children
Hindu
OLS
IVC
IVC+L
OLS
IVC
IVC+L
OLS
IVC
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
0.188∗∗∗ 2.573∗
2.567∗
(0.039)
(1.376)
(1.374)
−0.036∗∗∗−0.057∗∗∗ −0.057∗∗∗
(0.004)
(0.014)
(0.014)
−0.454∗∗∗−0.195
−0.196
(0.044)
(0.165)
(0.165)
0.084∗
0.068
0.068
(0.045)
(0.051)
(0.051)
0.162∗∗∗ 2.536∗
2.524∗
(0.037)
(1.374)
(1.369)
−0.024∗∗∗−0.043∗∗∗ −0.043∗∗∗
(0.004)
(0.013)
(0.013)
−0.561∗∗∗−0.313∗∗ −0.315∗∗
(0.043)
(0.159)
(0.158)
0.073∗
0.061
0.061
(0.040)
(0.047)
(0.047)
0.085
0.063
0.063
(0.065)
(0.076)
(0.076)
−0.111∗∗∗−0.221∗∗∗ −0.221∗∗∗
(0.024)
(0.070)
(0.069)
−0.126∗∗∗−0.080∗ −0.081∗
(0.032)
(0.048)
(0.048)
0.420∗∗∗ 0.406∗∗∗ 0.406∗∗∗
(0.031)
(0.036)
(0.036)
1.239∗∗∗ 1.079∗∗∗ 1.080∗∗∗
(0.080)
(0.134)
(0.133)
yes
yes
yes
yes
yes
yes
no
no
no
Home
Livestock
Agriculture
Low Skill
High Skill
Year F.E.
Village F.E.
Household F.E.
Observations
F-stat first stage
K-P F-stat (weak ident)
yes
yes
no
yes
yes
no
yes
yes
no
10,850
10,850
10,850
4.735
5.045
4.735
2.523
10,809
10,809
10,809
4.905
5.266
4.915
2.638
0.148∗∗∗
(0.049)
−0.024∗∗∗
(0.005)
−0.423∗∗∗
(0.056)
0.184
(0.115)
0.207∗∗∗
(0.077)
−0.043
(0.035)
−0.091∗∗
(0.039)
0.293∗∗∗
(0.038)
0.755∗∗∗
(0.095)
yes
no
yes
10,809
2.499∗
(1.333)
−0.036∗∗∗
(0.011)
−0.202
(0.150)
0.270∗
(0.139)
0.130
(0.122)
−0.170∗∗
(0.084)
−0.150∗∗
(0.059)
0.196∗∗∗
(0.072)
0.745∗∗∗
(0.132)
yes
no
yes
10,809
4.906
5.533
Notes: All estimations contain a constant. Standard errors in parentheses are clustered at the village level. ***
p<0.01, ** p<0.05, * p<0.1. Income represents total household income excluding agriculture expressed in per adult
equivalent terms. Instruments used are density of transmission cables (IVC ) and both the cable density and variation
in land elevation (IVC+L ). Hindu, Home and Livestock are dummy variables. Agriculture, Low Skill and High Skill
are occupations estimated with respect to a benchmark corresponding to any kind of home based activities, including
students and unemployed. The Quality takes value 0 for no electricity, 1 for high quality electricity supply and 0.5
for low quality electricity supply. The threshold for high quality power supply in 2005 is fixed at an average of 18
hours of power per day.
37
B
Appendix Figures
Figure B.1: Indian transmission and rail networks
Source: ESRI ArcGIS World package and geocommons.
38
Figure B.2: Indian transmission and road networks
Source: ESRI ArcGIS World package and geocommons.
39
Figure B.3: Household Electrification Rate vs Mean Income by State, 2005
1
Andhra
Bihar
Gujarat
Haryana
Himachal
Kerala
Maharashtra,
Madhya
Orissa
Punjab
Rajasthan
Tamil
Uttar
West
Assam
Jharkhand
.Uttarakhand
0
Chhattisgarh
S
1
.8
.6
.4
.2
8.6
8.8
9
9.2
9.4
9.6
State
2tate
.2 Bengal
Pradesh
Nadu
Pradesh
percentage
mean
Pradesh
Pradesh
Goa
income
of electrified households
State percentage of electrified households
.2
.4
.6
.8
Himachal Pradesh
Punjab
Tamil Nadu
Andhra Pradesh
Gujarat
Haryana
Kerala
Maharashtra, Goa
Madhya Pradesh
Chhattisgarh
Uttarakhand
Jharkhand
Rajasthan
Uttar Pradesh
West Bengal
Orissa
0
Bihar
Assam
8.6
8.8
9
9.2
State mean income
9.4
9.6
1
Andhra
Bihar
Gujarat
Haryana
Himachal
Kerala
Maharashtra,
Madhya
Orissa
Punjab
Rajasthan
Tamil
Uttar
West
Assam
Jharkhand
.Uttarakhand
0
Chhattisgarh
S
1
.8
.6
.4
.2
8.6
8.8
9
9.2
9.4
9.6
State
2tate
.2 Bengal
Pradesh
Nadu
Pradesh
mean
Pradesh
Pradesh
Goa
power quality
income
Himachal Pradesh
Kerala
State mean power quality
.4
.6
.8
Tamil Nadu
Gujarat
Andhra Pradesh
Maharashtra, Goa
Punjab
Haryana
Jharkhand
Chhattisgarh
Uttarakhand
Madhya Pradesh
Orissa
.2
West Bengal
Rajasthan
Uttar Pradesh
0
Bihar
Assam
8.6
8.8
9
9.2
State mean income
Note: A linear fit to the scatter plot is shown.
40
9.4
9.6