A Comprehensive Analysis of Poverty in India

This paper offers a comprehensive analysis of poverty in India. It shows that no matter which of the two official poverty lines is used, poverty has declined steadily in all states and for all social and religious groups. Accelerated growth between fiscal years 2004-2005 and 2009-2010 led to an accelerated decline in poverty rates. Moreover, the decline in poverty rates during these years was sharper for the socially disadvantaged groups relative to upper caste groups, so that a narrowing of the gap in the poverty rates is observed between the two sets of social groups. The paper also provides a discussion of the recent controversies in India regarding the choice of poverty lines.


Introduction
This paper provides comprehensive up-to-date estimates of poverty by social and religious groups in the rural and urban areas of the largest 17seventeen states in India. The specific measure of poverty reported in the paper is the poverty rate or head-countratio (HCR), which is the proportion of the population with expenditure or income below a pre-specified level referred to as the poverty line. In the context of most developing countries, the poverty line usually relates to a pre-specified basket of goods presumed to be necessary for above-subsistence existence.
In so far as prices vary across states and between rural and urban regions within the same state, the poverty line varies in nominal rupees across states and between urban and rural regions within the same state. 1 Similarly, since prices rise over time due to inflation, the poverty line in nominal rupees in a given location is also adjusted upwards over time.
The original official poverty estimates in India, provided by the Planning Commission, were based on the Lakdawala poverty lines so named after Professor D. T. Lakdawala who headed a 1993 expert group that recommended these lines. Recommendations of a 2009 expert committee headed by Professor Suresh Tendulkar led to an upward adjustment in the rural poverty line relative to its Lakdawala counterpart. Therefore, whereas the official estimates for earlier years are based on the lines and methodology recommended by the expert group headed by Lakdawala, those for more recent years have been based on the line and methodology recommended by the Tendulkar Committee. Official estimates based on both lines and methodologies exist for only two years, 1993-94 and 2004-05. These estimates are provided for the overall population, for rural and urban regions of each state and for the country as a whole. The Planning Commission does not provide estimates by social or religious groups.
In this paper, we provide estimates using both Lakdawala and Tendulkar lines for different social and religious groups in rural and urban areas in all major states and at the national level. Our estimates based on Lakdawala lines are computed for all years beginning in 1983 for which large or "thick" expenditure surveys have been conducted. Estimates based on the Tendulkar line and methodology are provided for the three latest large expenditure surveys, 1993-94, 2004-05 and 2009-10 2 . Our objective in writing the paper is twofold. First, much confusion has arisen in the policy debates in India around issues such as whether or not growth has helped the poor; if yes, how much and over which time period; and whether growth is leaving certain social or religious groups behind. We hope that by providing poverty estimates for various time periods, social groups, religious groups, states and urban and rural areas in one place this paper will help ensure that future policy debates are based on fact. Second, researchers interested in explaining how various policy measures impact poverty 1 Prices could vary not just between urban and rural regions within a state but also across sub-regions within rural and sub-regions within urban regions of a state. Therefore, in principle, we could envision many different poverty lines within rural and within urban regions in each state. To keep the analysis manageable, we do not make such finer distinctions in the paper. 2 Panagariya and More (2013) use data from the 68 th Round of the national Sample Survey and provide results for 2011-2012. 3

The Expenditure Surveys
The main source of data for estimating poverty in India is the expenditure survey conducted by the National Sample Survey Office. India is perhaps the only developing country that began conducting such surveys on a regular basis as early as 1950-51. The surveys have been conducted at least once a year since 1950-51 though the sample was too small to permit reliable estimates of poverty at the level of the state until 1973-74. A decision was made in the early 1970s to replace the smaller annual surveys by large-size expenditure (and employment-unemployment) surveys to be conducted every five years.
While the NSSO collects the data and produces reports providing information on monthly per-capita expenditures and their distribution in rural and urban areas of different states and at the national level, it is the Planning Commission that computes the poverty lines and provides official estimates of poverty. The official estimates are strictly limited to quinquennial surveys and to rural, urban and total populations in different states and at the national level. The official estimates are not provided for specific social or religious groups. These can be calculated selectively for specific groups or specific years by researchers. With rare exceptions, discussions and debates on poverty have been framed around the quinquennial surveys even though the non-quinquennial survey samples are large enough to allow reliable estimates at the national level.
For each household interviewed, the survey collects data on the quantity of and expenditure on a large number of items purchased. For items such as education and health services for which the quantity cannot be meaningfully defined, only expenditure data are collected. The list of items is elaborate. For example, the 66 th round collected data on 142 items of food, 15 items of energy, 28 items of clothing, bedding and footwear, 19 items of educational and medical expenses, 51 items of durable goods, and 89 other items.
It turns out that household responses vary systematically according to the length of the reference period to which the expenditures are related. For example, a household could be asked about its expenditures on durable goods during the preceding 30 days or the entire year. When the information provided in the first case is converted into annual expenditure, it is found to be systematically lower than when the survey directly asks households to report their annual expenditures. Therefore, estimates of poverty vary depending on the reference period chosen in the questionnaire.
Most quinquennial surveys have collected information on certain categories of relatively infrequently purchased items including clothing and consumer durables on the basis of both 30-days and 365-days reference periods. For other categories including all food and fuel and consumer services, they have used a 30-days reference period. The data allow us to estimate two alternative measures of monthly per-capita expenditures: • Uniform Reference Period (URP): All expenditure data used to estimate monthly per-capita expenditure are based on the 30-days reference period. • Mixed Reference Period (MRP): Expenditure data used to estimate the monthly per-capita expenditure are based on the 365-days reference period in the case of clothing and consumer durables and the 30-days reference period in the case of other items. With rare exceptions, monthly per-capita expenditure associated with the MRP turns out to be higher than that associated with the URP. The original Planning Commission estimate of poverty, which had employed the Lakdawala poverty lines, had relied on the URP monthly per-capita expenditures. At some time prior to the Tendulkar Committee report, the Planning Commission decided, however, to shift to the MRP estimates. Therefore, while recommending revisions that led to an upward adjustment in the rural poverty line, the Tendulkar Committee also shifted to the MRP monthly percapita expenditures in its poverty calculations. Therefore, the revised poverty estimates available for 1993-94, 2004-05 and 2009-10 are based on the Tendulkar lines and the MRP estimates of monthly per-capita expenditures.

The NSSO versus NAS Expenditure Estimates
We note an important feature of the NSSO expenditure surveys at the outset. The average monthly per-capita expenditure based on the surveys falls well short of the average private consumption expenditure separately available from the National Accounts Statistics (NAS) of the Central Statistical Office (CSO). Moreover, the proportionate shortfall has been progressively rising over successive surveys. These two observations hold regardless of whether we use the URP or MRP estimate of monthly per-capita expenditure available from the NSSO. Figure 1 graphically depicts this phenomenon in the case of URP monthly per-capita expenditure, which is more readily available for all quinquennial surveys since 1983.
Precisely what explains the gap between the NSSO and NAS expenditures has important implications for poverty estimates. For example, if the gap in any given year is uniformly distributed across all expenditure classes as Bhalla (2002) assumes in his work, true expenditure in 2009-10 is uniformly more than twice what the survey finds. This would imply that many individuals currently classified as below the poverty line are actually above it. Moreover, a recognition that the proportionate gap between NSSO and NAS private expenditures has been rising over time implies that the poverty ratio is being over-estimated by progressively larger margins over time. At the other extreme, if the gap between NSSO and NAS expenditures is explained entirely by under-reporting of the expenditures by households classified as non-poor, poverty levels will not be biased upwards.
There are good reasons to believe, however, that the truth lies somewhere between these two extremes. The survey underrepresents wealthy consumers. For instance, it is unlikely that any of the billionaires and most of the millionaires are covered by the survey. Likewise, the total absence of error among households below the poverty line is highly unlikely. For example, recall that the expenditures on durables are systematically under-reported for the 30-days reference period relative to that for 365days reference period. Thus, in all probability households classified as poor account for a 5 part of the gap so that there is some over-estimation of the poverty ratio at any given poverty line. 3

The Official Poverty Lines
The 1993 expert group headed by Lakdawala defined all-India rural and urban poverty lines in terms of per-capita total consumption expenditure at 1973-74 market prices. The underlying poverty-line-consumption baskets were anchored in the per-capita calorie norms of 2400 and 2100 in rural and urban areas, respectively. They also provided for the consumption of all goods and services present in the rural and urban poverty line baskets. The lines were based on different underlying baskets, however. This meant that the two poverty lines represented different levels of real expenditures.
State-level rural poverty lines were derived from the national rural poverty line by adjusting the latter for price differences between national and state-level consumer price indices for agricultural laborers. Likewise, state-level urban poverty lines were derived from the national urban poverty line by adjusting the latter for price differences between the national and state-level consumer price indices for industrial laborers. National and state-level rural poverty lines were adjusted over time by applying the national and statelevel price indices for agricultural workers, respectively. Urban poverty lines were adjusted similarly over time.
Lakdawala lines served as the official poverty lines until 2004-05. The Planning Commission applied them to URP-based expenditures in the quinquennial surveys to calculate official poverty ratios. Criticisms of these estimates on various grounds led the Planning Commission to appoint an expert group under the chairmanship of Suresh Tendulkar in December 2005 with the charge to recommend appropriate changes in methodology to compute poverty estimates. The group submitted its report in 2009.
In its report, the Tendulkar committee (Planning Commission 2009) noted three deficiencies of the Lakdawala poverty lines. First, the poverty line baskets remained tied to consumption patterns observed in 1973-74. But more than three decades later, these baskets had shifted, even for the poor. Second, the consumer price index for agricultural workers understated the true price increase. This meant that over time, the upward adjustment in the rural poverty lines was less than necessary so that the estimated poverty ratios understated rural poverty. Finally, the assumption that health and education would be largely provided by the government, underlying Lakdawala lines, did not hold any longer. Private expenditures on these services had risen considerably, even for the poor. This change was not adequately reflected in the Lakdawala poverty lines.
To remedy these deficiencies, the Tendulkar committee began by noting that the NSSO had already decided to shift from URP-based expenditures to MRP-based expenditures to measure poverty. With this in view, the committee's first step was to situate the revised poverty lines in terms of MRP expenditures in some generally acceptable aspect of the existing practice. To this end, it observed that since the nationwide urban poverty ratio of 25.7 percent, calculated from URP-based expenditures in the 2004-05 survey, was broadly accepted as a good approximation of prevailing urban poverty, the revised urban poverty line should be anchored to yield this same estimate using MRP-based per-capita consumption expenditure from the 2004-05 survey. This decision led to the MRP-based per-capita expenditure of the individual at the 25.7 percentile in the national distribution of per-capita MRP expenditures as the national urban poverty line.
The Tendulkar committee further argued that the consumption basket associated with the national urban poverty line also be accepted as the rural poverty line consumption basket. This implied the translation of the new urban poverty line using the appropriate price index to obtain the nationwide rural poverty line. Under this approach, rural and urban poverty lines became fully aligned. Applying MRP-based expenditures, the new rural poverty line yielded a rural poverty ratio of 41.8 percent in 2004-05 compared with 28.3 percent under the old methodology. State-level rural and urban poverty lines were also to be derived from the national urban poverty line by applying the 7 appropriate price indices derived from the price information within the sample surveys themselves. This methodology fully aligned all poverty lines.

Controversies Regarding Poverty Lines 4
We address here the two rounds of controversies over the poverty line that broke out in the media in September 2011 and March 2012. The first round of controversy began with the Planning Commission filing an affidavit with the Supreme Court stating that the poverty line at the time was on average 32 and 26 rupees per person per day in urban and rural India, respectively. Being based on the Tendulkar methodology, these lines were actually higher than the Lakdawala lines on which the official poverty estimates had been based until 2004-05. However, the media and civil society groups pounced on the Planning Commission for diluting the poverty lines so as to inflate poverty reduction numbers and to deprive many potential beneficiaries of entitlements. For its part, the Planning Commission did a poor job of explaining to the public precisely what it had done and why.
The controversy resurfaced in March 2012 when the Planning Commission released the poverty estimates based on the 2009-10 expenditure survey. The Planning Commission reported that these estimates were based on average poverty lines of 28.26 and 22.2 rupees per person per day in urban and rural areas, respectively. Comparing these lines to those previously reported to the Supreme Court, the media once again accused the Planning Commission of lowering the poverty lines. 5 The truth of the matter was that whereas the poverty lines reported to the Supreme Court were meant to reflect the price level prevailing in mid-2011, those underlying poverty estimates for 2009-10 were based on the mid-point of 2009-10. The latter poverty lines were lower because the price level at the mid-point of 2009-10 was lower than that in mid-2011. In real terms, the two sets of poverty lines were identical.
While there was no basis to the accusations that the Planning Commission had lowered the poverty lines, the issue of whether the poverty lines remain excessively low despite having been raised does require further examination. In addressing this issue, it is important to be clear about the objectives behind the poverty line.
Potentially, there are two main objectives behind poverty lines: to track the progress made in combating poverty and to identify the poor towards whom redistribution programs can be directed. The level of the poverty line must be evaluated separately against each objective. In principle, we may want separate poverty lines for the two objectives.
With regard to the first objective, the poverty line should be set at a level that allows us to track the progress made in helping the truly destitute or those living in abject poverty, often referred to as extreme poverty. Much of the media debate during the two episodes focused on what could or could not be bought with the poverty-line 8 expenditure. 6 There was no mention of the basket of goods that was used by the Tendulkar Committee to define the poverty line.
In Annexure E of its report, the Tendulkar Committee gave a detailed itemized list of the expenditures of those "around poverty line class for urban areas in all India". Unfortunately, it did not report the corresponding quantities purchased of various commodities. In this paper, we now compute these quantities from unit-level data where feasible and report them in Table 1 for a household consisting of five members. 7 Our implicit per-person expenditures on individual items are within 3 rupees of their corresponding expenditures reported in Annexure E of the Tendulkar Committee report.
We report quantities wherever the relevant data are available. In the survey, the quantities are not always reported in weights. For example, lemons and oranges are reported in numbers and not in kilograms. In these cases, we have converted the quantities into kilograms using the appropriate conversion factors. The main point to note is that while the quantities associated with the poverty line basket may not permit a comfortable existence including a balanced diet, they allow above-subsistence existence. The consumption of cereals and pulses at 50.9 and 3.5 kilograms compare with 48 and 5.5 kilograms, respectively, for the mean consumption of the top 30 percent of the population. Likewise the consumption of edible oils and vegetables at 2.7 and 23.9 kilograms for the poor compare with 4.5 and 35.5 kilograms, respectively, for the top 30 percent of the population. 8 This comparison shows that at least in terms of the provision of two square meals a day, the poverty line consumption basket is compatible with above-subsistence-level consumption.
We reiterate our point as follows. In 2009-10, the urban poverty line in Delhi was 1040.3 rupees per person per month (34.2 rupees per day). For a family of five, this amount would translate into 5,201.5 rupees per month. Assuming that each family member consumes ten kilograms per month of cereal and one kilogram per month of pulses and the prices of the two grains are 15 and 80 rupees per kilogram, respectively, the total expenditure on grain would be 1,150 rupees. 9 This would leave 4,051.5 rupees for milk, edible oils, fuel, clothing, rent, education, health and other expenditures. While this amount may not allow a fully balanced diet, comfortable living and access to good education and health, it is consistent with an above-subsistence level of existence. Additionally, if we take into account access to public education and health and subsidized grain and fuel from the public distribution system, the poverty line is scarcely out of line with the one that would allow exit from extreme poverty.
But what about the role of the poverty line in identifying the poor for purposes of redistribution? Ideally, this exercise should be carried out at the local level in light of resources available for redistribution since the poor must ultimately be identified locally. 6 For instance, one commentator argued in a heated television debate that since bananas in Jor Bagh (an upmarket part of Delhi) cost Rupees 60 a dozen, an individual could barely afford two bananas per meal per day at poverty line expenditure of 32 rupees per person per day. 7 We thank Rahul Ahluwalia for supplying us with Table 1 Nevertheless, if the national poverty line is used to identify the poor, could we still defend the Tendulkar line as adequate? We argue in the affirmative. Going by the urban and rural population weights of 0.298 and 0.702 implicit in the population projections for January 1, 2010, the average countrywide per-capita MRP expenditure during 2009-10 works out to 40.2 rupees per person per day. Therefore, going by the expenditure survey data, equal distribution across the entire country would allow barely 40.2 rupees per person per day in expenditures. Raising the poverty line significantly above the current level must confront this limit with regard to the scope for redistribution.
It could be argued that this discussion is based on the expenditure data in the expenditure survey, which underestimates true expenditures. The scope for redistribution might be significantly greater if we go by expenditures as measured in the National Accounts Statistics. The response to this criticism is that the surveys underestimate not just the average national expenditure, but also the expenditures of those identified as poor. Depending on the extent of this underestimation, the need for redistribution itself would be overestimated.
Even so, it is useful to test the limits of redistribution by considering the average expenditure according to the National Accounts Statistics. The total private final consumption expenditure at current prices in 2009-10 was 37,959.01 billion rupees. Applying the population figure of 1.174 billion as of January 1, 2010 in the NSSO 2009-10 expenditure survey, this total annual expenditure translates into daily expenditure of 88.58 rupees per person. This figure includes certain items such as imputed rent on owner occupied housing and expenditures other than those by households such as the expenditures of civil society groups, which would not be available for redistribution. Thus, per-capita expenditures achievable through equal distribution, even when we consider the expenditures as per the national accounts statistics, is likely to by modest.
To appreciate further the folly of setting too high a poverty line for purposes of identifying the poor, recall that the national average poverty line was 22.2 rupees per person per day in rural areas and 28.26 rupees in urban areas in 2009-10. Going by the expenditure estimates for different expenditure classes in Government of India (2011a), raising these lines to just 33.3 and 45.4 rupees, respectively, would place 70% of the rural and 50% of the urban population in poverty in 2009-10. If we went a little further and set the rural poverty line at 39 rupees per day and the urban poverty line at 81 rupees per day in 2009-10, we would place 80 percent of the population in each region below the poverty line. Will the fate of the destitute not be compromised if the meager tax revenues available for redistribution were thinly spread on this much larger population? 10

Poverty at the National Level
Official poverty estimates are available at the national and state levels for the entire population but not by social or religious groups for all years during which the NSSO conducted quinquennial surveys. Excluding the 1999-2000 survey, which became non-comparable to other quinquennial surveys due to a change in sample design, these years consist of 1973-74, 1977-78, 1983, 1987-88, 1993-94, 2004-05 and 2009-10 11 . The Planning Commission has published poverty ratios for the first six of these surveys at the 11 Lakdawala lines and for the last three at the Tendulkar lines for rural and urban areas at the national and state levels.
In this paper, we provide comparable poverty rates for all of the last five quinquennial surveys including 2009-10 at Lakdawala lines. For this purpose, we update the 2004-05 Lakdawala lines to 2009-10 using the price indices implicit in the official Tendulkar lines for 2004-05 and 2009-10 at the national and state levels. We provide estimates by both social and religious groups for all quinquennial surveys beginning in 1983 at the Lakdawala lines and for the years relating to the last three such surveys at Tendulkar lines at the national and state levels.
While we focus mainly on the evolution of poverty since 1983 in this paper, it is useful to begin with a brief look at the poverty profile in the early years. This is done in Figure 2 using the estimates in Datt (1998Datt ( ) for years 1951Datt ( -52 to 1973 The key message of the graph is that the poverty ratio hovered between approximately 50 and 60 percent with a mildly rising trend. This is not surprising. India was extremely poor at independence. Subsequently, unlike countries such as Taiwan, China; South Korea; Singapore; and Hong Kong SAR, China, the country grew very slowly. Growth in percapita income during these years was a mere 1.5 percent per year. Such low growth coupled with a very low starting per-capita income meant at best limited scope for achieving poverty reduction even through redistribution. As argued above, even today, after more than two decades of almost 5 percent growth in per-capita income, the scope for redistribution remains limited. 12 Figure 2: The poverty ratio in India, 1951India, -52 to 1973 We are now in a position to provide the poverty rates for the major social groups based on the quinquennial expenditure surveys beginning in 1983. The social groups identified in the surveys are Scheduled Castes (SC), Scheduled Tribes (ST), Other Backward Castes (OBC) and the rest, which we refer to as forward castes (FC). In addition, we define the non-scheduled castes as consisting of the OBC and FC. The NSSO began identifying the OBC beginning in 1999-2000. Since we are excluding this survey due to its lack of comparability with other surveys, the OBC as a separate group begins appearing in our estimates from 2004-05 only.
In Table 2, we provide the poverty rates at the Lakdawala lines in rural and urban areas and the two regions combined at the national level. Four features of this table are worthy of note. First and foremost, the poverty rates have declined between every pair of successive surveys for every single social group in each rural and urban area. Contrary to common claims, growth has been steadily helping the poor from every broad social group rather than leaving the socially disadvantaged behind. Second, predictably, the poverty rates in rural India are consistently the highest for the ST followed by the SC, OBC and FC in that order. This pattern also holds in urban areas though with some exceptions. In particular, in some years, the ST poverty 13 rates are lower than the SC rates but this is not of great significance since more than 90 percent of the ST population lives in rural areas.
Third, with growth accelerating to above 8 percent beginning in 2003-04, poverty reduction between 2004-05 and 2009-10 also accelerated. The percentage-point reduction during this period was larger than during any other five-year period. Most importantly, the acceleration was the greatest for the ST and SC in that order so that at last the gap in poverty rates between the scheduled and non-scheduled groups declined significantly.
Finally, while the rural poverty rates were slightly higher than the urban rates for all groups in 1983, the order switched for one or more groups in several of the subsequent years. Indeed, in 2009-10, the urban rates turn out to be uniformly higher for every single group. This largely reflects progressive misalignment of the rural and urban poverty lines with the former becoming lower than the latter. It was this misalignment that led the Tendulkar Committee to revise the rural poverty line to realign it to the higher, urban line. Table 3 reports the poverty estimates based on the Tendulkar lines. Recall that the Tendulkar line holds the urban poverty ratio at 25.7 percent in 2004-05 when measuring poverty at MRP expenditures. Our urban poverty ratio in Table 3 reproduces this estimate within 0.1 percentage point.
The decline in poverty rates between every two successive surveys for every social group in rural as well as urban areas, which we noted at the Lakdawala lines in Table 2, remains valid at the Tendulkar lines. Moreover, rural poverty ratios now turn out to be higher than their urban counterparts for each group in each year. As in Table 2, the decline is the sharpest during the high-growth period between 2004-05 and 2009-10. Finally and most importantly, the largest percentage-point decline between these years in rural and urban areas combined is for the ST followed by the SC, OBC and FC in that order. Given that the ST also had the highest poverty rates followed by SC, OBC and ST in that order in 2004-05, this pattern implies that the socially disadvantaged groups have done significant catching up with the better off groups. This is a major break with past trends 13 . Next, we report the national poverty rates by religious groups. In Table 4, we show the poverty rates at Lakdawala lines in rural and urban India and the country taken as a whole. Three observations follow. First, at the aggregate level (rural plus urban), poverty rates show a decline between every pair of successive surveys in the case of Hindus, Muslims, Christians, Jains and Sikhs. Poverty among the Buddhists also declines steadily with the exception of between 1983 and 1987-88. With one exception (Muslims in rural India between 1987-88 and 1993-94), the pattern of declining poverty rates between any two successive surveys also extends to the rural and urban poverty rates in the case of the two largest religious communities, Hindus and Muslims.
Second, going by the poverty rates in 2009-10 in rural and urban areas combined, Jains have the lowest poverty rates followed by Sikhs, Christians, Hindus, Muslims and Buddhists in that order. Prosperity among Jains and Sikhs is well known but the lower level of poverty among Christians relative to Hindus is less well known. Also interesting is the relatively small gap of 5.8 percentage points between poverty rates among Hindus and Muslims.
Finally, the impact of accelerated growth on poverty between 2004-05 and 2009-10 that we observed across social groups can also be seen across religious groups. Once again, we see a sharper decline in the poverty rate for the largest minority, Muslims, relative to Hindus who form the majority of the population.
This broad pattern holds when we consider poverty rates by religious groups at the Tendulkar line, as seen in Table 5. Jains have the lowest poverty rates followed by Sikhs, Christians, Hindus, Muslims and Buddhists in that order. With one exception (Sikhs in rural India between 1993-94 and 2004-05), poverty declines between every pair of successive surveys for every religious group in rural as well as urban India. The only difference is that the decline in poverty among Muslims in rural and urban areas combined between 2004-05 and 2009-10 is not as sharp as at the Lakdawala lines 14 . As a result, we do not see a narrowing of the difference in poverty between Hindus and Muslims when taking rural and urban regions together. We do see a narrowing of the difference in urban poverty but this gain is neutralized by the opposite movement in the rural areas due to a very sharp decline in poverty among Hindus, perhaps due to the rapid decline in poverty among the SC and the ST.

Poverty in the States: Rural and Urban
We now turn to the progress made in poverty alleviation in different states. Though our focus in the paper is on poverty by social and religious groups, we first consider it at the aggregate level in rural and urban areas. India has 28 states and 7 union territories. To keep the analysis manageable, we limit ourselves to the 17 largest states. 15 Together, these states account for 95 percent of the total population. We exclude all seven union territories including Delhi; the smallest six of the seven northeastern states (retaining only Assam); and the states of Sikkim, Goa, Himachal Pradesh and Uttaranchal. Going by the expenditure survey of 2009-10, each of the included states has a population exceeding 20 million while each of the excluded states has a population less than 10 million. Among the union territories, only Delhi has a population exceeding 10 million.

Rural and Urban Populations
We begin by presenting, in Table 6, the total population in each of the 17 largest states and its distribution between rural and urban areas as revealed by the NSSO expenditure survey of 2009-10. 16 The population totals in the expenditure survey are lower than the corresponding population projections by the Registrar General and Census Commissioner of India (2006) as well as those implied by Census 2011. 17 Our choice is dictated by the fact that poverty estimates should be evaluated with reference to the population underlying the survey design instead of those suggested by external sources. For example, the urban poverty estimate in Kerala in 2009-10 must be related to the urban population in the state underlying the expenditure survey in 2009-10 instead of projections based on the Census 2001 and Census 2011. 18 According to Table 6, 27 percent of the national population lived in urban areas and the remaining 73 percent in the rural areas in 2009-10. This composition understates the true share of the urban population, which was revealed to be 31.2 percent in the Census 2011. The table shows ten states having populations of more than 50 million (60 million according to the Census 2011). We will refer to these ten states as the large 15 Although Delhi has its own elected legislature and Chief Minister, it remains a union territory. For example, central home ministry has the effective control of the Delhi police through lieutenant governor who is the de jure head of the Delhi government and appointed by the Government of India. 16 Our absolute totals for rural and urban areas of the states and India in Table 6 match those in Tables 1A-R and 1A-U, respectively, in Government of India (2011b). 17 The Planning Commission derives the absolute number of poor from poverty ratios using census-based population projections. Therefore, the population figure underlying the absolute number of poor estimated by the Planning Commission are higher than those in Table 6, which are based on the expenditure survey of 2009-10. 18 This distinction is a substantive in the case of states in which the Censuses reveal the degree of urbanization to be very different than that underlying the design of the expenditure surveys. For example, the expenditure survey of 2009-10 places the urban population in Kerala at 26 percent of the total in 2009-10. But the Census 2011 finds the rate of urbanization in the state to be 47.7 percent. states. They account for a little more than three-fourths of the total population of India. At the other extreme, eleven small states (excluded from our analysis and therefore not shown in Table 6) have populations of less than ten million (13 million according to the Census 2011) each. The remaining seven states, which we call medium-size states, have populations ranging from 36 million in Orissa to 22 million in Chhattisgarh (42 million in Orissa to 25.4 million in Chhattisgarh, according to the Census 2011). Among the large states, Tamil Nadu, Maharashtra, Gujarat and Karnataka, in that order, are the most urbanized with 35 percent or higher rate of urbanization. Bihar is the least urbanized among the large states and has an urbanization rate of just 10 percent. Among the medium-size states, only Punjab has 35 percent urban population, with the rest having urbanization rates of 30 percent or less. Assam and Orissa, with just 10 and 14 percent urban populations, respectively, are the least urbanized medium-size states.

Rural and Urban Poverty
We now turn to the estimates of rural and urban poverty in the 17 largest states. To conserve space, throughout the text of the paper, we confine ourselves to presenting the estimates at the Tendulkar line. In the appendix, we report the estimates at the Lakdawala lines. Recall that the estimates at the Tendulkar line are available for three years: 1993years: -94, 2004years: -05 and 2009years: -10. Disregarding 1973years: -74 and 1977, which are outside the cope of our paper, estimates at Lakdawala lines are available for two additional years: 1983 and 1987-88. Table 7 reports the poverty estimates with the states arranged in descending order of their populations. Several observations follow. First, taken as a whole, poverty fell in each of the 17 states between 1993-94 and 2009-10. When we disaggregate rural and urban areas within each state, we still find a decline in poverty in all states in each region over this period. Indeed, if we take the ten largest states, which account for three-fourths of India's population, every state except Madhya Pradesh experienced a decline in both rural and urban poverty between every two successive surveys. The reduction in poverty with rising incomes is a steady and nationwide phenomenon and not driven by the gains made in a few specific states or just rural or just urban areas of a given state.
Second  20 Finally, among the large states, Tamil Nadu has the lowest poverty ratio followed by Andhra Pradesh and Gujarat in that order. Tamil Nadu, Karnataka and Andhra Pradesh-all of them from the south-have made the largest percentage-point gains in poverty reduction among the large states between 1993-94 and 2009-10. Among the medium-size states, Kerala and Haryana in that order have the lowest poverty rates while Orissa and Jharkhand have made the largest percentage-point gains during 1993-94 to 2009-10.
It is useful to relate the poverty levels to per-capita expenditures. In Table 8, we present per-capita expenditures in current rupees in the 17 states in the three years for which we have the poverty ratios, with the states ranked in descending order of population. Ideally, we should have the MRP expenditures for all three years but since they are available for only the last two years, we report the URP expenditures for 1993-94. Several observations follow from a comparison of Tables 7 and 8.  Tamil Nadu  294  438  602  1166  1017  1795  Madhya Pradesh  252  408  461  893  803  1530  Rajasthan  322  425  598  945  1035  1577  Gujarat  303  454  645  1206  1065  1914  Karnataka  269  423  543  1138  888  2060  Orissa  220  403  422  790  716  1469  Kerala  390  494  1031  1354  1763  2267  Assam  258  459  577  1130  867  1604  Jharkhand  439  1017  724  1442  Haryana  385  474  905  1184  1423  2008  Punjab  433  511  905  1306  1566  2072  Chhattisgarh  445  963  686  1370  All-India  281  458  579  1105  953  1856 First, high per-capita expenditures are associated with low poverty ratios. For example consider rural poverty in 2009-10. Kerala, Punjab and Haryana in that order have the highest rural per-capita expenditures. They also have the lowest poverty ratios in the same order. At the other extreme, Chhattisgarh and Bihar in that order have the lowest rural per-capita expenditures and also the highest rural poverty ratios. More broadly, the top nine states by rural per-capita expenditure are also the top nine states in terms of low poverty ratios. A similar pattern can also be found for urban per-capita expenditures and urban poverty. Once again, Kerala ranks at the top and Bihar at the 21 bottom in terms of each indicator. Figure 3 offers a graphical representation of the relationship in rural and urban India in 2009-10 using state level data. Second, one state, which stands out in terms of low poverty ratios despite a relatively modest ranking in terms of per-capita expenditure, is Tamil Nadu. It ranked eighth in terms of rural per-capita expenditure but fourth in terms of rural poverty in 2009-10. In terms of urban poverty it did even better, ranking a close second despite its ninth rank in terms of urban per-capita expenditure. Gujarat also did very well in terms of urban poverty, ranking third in spite of the seventh rank in terms of urban per-capita expenditure.
Finally, there is widespread belief that Kerala has achieved the lowest rate of poverty despite its low per-capita income through more effective redistribution. Table 8 entirely repudiates this thesis. In 1993-94, Kerala already had the lowest rural and urban poverty ratios and it enjoyed the second highest rural per-capita expenditure and third highest urban per-capita expenditure among the 17 states. Moreover, in terms of percentage-point reduction in poverty, all other southern states dominate Kerala. For example, between 1993-94 and 2004-05, Tamil Nadu achieved 27.4-percentage points reduction in poverty compared to 19.3 percentage points by Kerala. We may also add that Kerala has had very high inequality of expenditures. In 2009-10, the Gini coefficient associated with expenditures in the state was by far the highest among all states in rural as well as urban areas.

Poverty in the States by Social Groups
In this section we decompose population and poverty by social groups. As previously mentioned, traditionally, the expenditure surveys have identified the social group of the households using a three-way classification: Scheduled Castes, Scheduled Tribes and non-scheduled castes. But beginning with the 1999-2000 survey, the last category was further subdivided into Other Backward Castes and the rest. The latter is sometimes called forward castes, a label we use in this paper.
We begin by describing the shares of the four social groups in the total population of the 17 states. Table 9 reports the shares of various social groups in the 17 largest states according to the expenditure survey of 2009-10. We continue to rank the states according to population from the largest to the smallest. Nationally, the Scheduled Tribes constitute 9 percent of the total population of India according to the expenditure survey of 2009-10. In past surveys and the Census 2001, this proportion was 8 percent. The Scheduled Castes form 20 percent of the total population according to the NSSO expenditure surveys, though the Census 2001 placed this proportion at 16 percent. The OBC are not identified as a separate group in the censuses so that their proportion is available from the NSSO surveys only. The figure has varied from 36 to 42 percent across the three quinquennial expenditure surveys since the OBC began to be recorded as a separate group.

Population Distribution by Social Groups within the States
The Scheduled Tribes are more unevenly divided across states than the remaining social groups. In so far as the ST were very poor at independence and they happen to be outside the mainstream of the economy, ceteris paribus, states with high proportions of ST population are at a disadvantage in combating poverty relative to the other states. From this perspective, the four southern states enjoy a clear advantage: Kerala and Tamil Nadu have virtually no tribal populations while Andhra Pradesh and Karnataka have proportionately smaller tribal populations (5 and 9 percent of the total, respectively) than some of the northern states with high concentrations.
Among the large states, Madhya Pradesh, Gujarat and Rajasthan in that order have proportionately the largest concentrations of ST populations. The ST constitute 20, 17 and 14 percent of their respective populations. Some of the medium-size states, of course, have proportionately even larger concentrations of the ST populations. These include Chhattisgarh, Jharkhand and Orissa with the ST forming 30, 29 and 22 percent of their populations, respectively.
Since the traditional exclusion of the SC has meant that they began with a very high incidence of abject poverty and low levels of literacy, states with high proportions of them also face an uphill task in combating poverty. Even so, since the SC populations are not physically isolated from the mainstream of the economy, there is greater potential for the benefits of growth reaching them than the ST. This is illustrated, for example, by the emergence of some rupee millionaires among the SC but not the ST during the recent high-growth phase (Dehejia and Panagariya 2012).
Once again, at 9 percent, Kerala happens to have proportionately the smallest SC population among the 17 states listed in Table 9. Among the largest 10 states, West Bengal, Uttar Pradesh, Bihar, Rajasthan and Madhya Pradesh in that order have the highest concentrations of the SC populations. Among the medium-size states, Punjab, Haryana and Orissa in that order have proportionately the largest SC populations.
The SC and ST populations together account for as much as 40 and 35 percent of the total state population in Madhya Pradesh and Rajasthan, respectively. At the other extreme, in Kerala these groups together account for only 10 percent of the population. These differences mean that ceteris paribus, Madhya Pradesh and Rajasthan face a significantly more uphill battle in combating poverty than Kerala.
The ST populations also differ from the SC in that they are far more heavily concentrated in rural than urban areas. Table 10   An implication of the small ST population in the urban areas in all states and in both rural and urban areas in a large number of states is that the random selection of households results in a relatively small number of ST households being sampled in urban areas nearly everywhere and in both rural and urban areas in many states. The problem is especially severe in many of the smallest states in which the total sample size is small in the first place. A small ST sample translates into a large error in the associated estimate of the poverty ratio. We will present the poverty estimates in all states and regions as long as positive group is sampled. Nevertheless, we caution the reader to the possibility of errors in Table 11 associated with the number of ST households in the 2009-10 survey.

Poverty by Social Groups
We now turn to poverty estimates by social groups. We present statewide poverty ratios at the Tendulkar line for the ST, SC and non-scheduled castes in Table 12 and for the OBC and FC in Table 13. Separate rural and urban poverty estimates at both the Tendulkar and Lakdawala lines for each group are relegated to the appendix. As before, we arrange the states from the largest to the smallest according to population in Tables 12  and 13.
With one exception, Chhattisgarh, the poverty ratio declines for each group in each state between 1993-93 and 2009-10. There is little doubt that rising incomes have helped all social groups nearly everywhere. In the vast majority of the states, we also observe acceleration in the decline in poverty between 2004-05 and 2009-10 compared to between 1993-93 and 2004-05. Reassuringly, the decline in ST and SC poverty has accelerated recently with the gap in poverty rates between them and the non-scheduled castes narrowing. The negative relationship between poverty ratios and per-capita expenditures we depicted in Figure 3 can also be observed for the social groups taken separately. Using the rural poverty estimates by social groups from the appendix, we show this relationship between SC poverty and per capita rural expenditures in the left panel and that between the ST poverty and per capita rural expenditures in the right panel of in Figure 4. Figure  4 closely resembles Figure 3. The fit in the right panel is poorer than that in the left panel as well as those in Figure 3. This is partially because the ST are often outside the mainstream of the economy and less responsive to rising per-capita incomes. This factor is presumably exacerbated by the fact that the number of observations in the case of the ST has been reduced to 11 due to the number of ST households in the sample dropping below 100 in six of the 17 states.  Table 13. Taking the estimates in Tables 12 and 13 together, it can be seen that on average, poverty rates are the highest for the ST followed by SC, OBC and FC in that order. At the level of the individual states, the ranking between SC and ST poverty rates is not clear-cut but with rare exceptions the poverty rates for these two groups exceed systematically those for the OBC, which in turn exceed the rates for the FC. An interesting feature of the FC poverty rates is their low level in all but a handful of the states. For example, in 2009-10, the FC poverty rate is just 3.9 percent in Punjab, 5.9 percent in Kerala, 6.5 percent in Haryana, 6.9 percent in Tamil Nadu and 10.5 percent even in Rajasthan. In 14 of the largest 17 states, the FC poverty rate is below 25 percent. The states with low FC poverty rates generally also have low OBC poverty rates making the proportion of the SC and ST population the key determinant of the state-wide rate.
This point is best illustrated by a comparison of poverty rates between Punjab and Kerala. Poverty rates for the non-scheduled caste population in 2009-10 is 7.3 percent in Punjab and 10.4 percent in Kerala while those for scheduled castes are 29.2 and 27.4 percent, respectively, in the two states. Yet, since the SC constitute 39 percent of the population in Punjab but only 9 percent in Kerala, the statewide poverty rate turns out to be 15.8 percent in the former and 12 percent for the latter.
The caste composition also partially helps explain the differences in the poverty rates between Maharashtra and Gujarat on the one hand and Kerala on the other. Statewide poverty rates in the former states were 24.8 and 23.2 percent, respectively, and 12 percent in 2009-10 in the latter (Table 10). In part, the differences follow from the significantly higher per-capita expenditures in Kerala, as seen from Table 11 19 . But Maharashtra and Gujarat also face a more uphill task of combating poverty on account of significantly higher proportions of the ST and SC populations. These groups respectively account for 17 and 11 percent of the total population in Gujarat and 10 and 15 percent in Maharashtra. In comparison, only 1 percent of the population is ST and 9 percent SC in Kerala (Table 9).

Poverty in the States by Religious Groups
We finally turn to poverty estimates by religious groups in the states. India is home to many different religious communities including Hindus, Muslims, Christians, Sikhism, Jains and Zoroastrians. Additionally, tribes follow their own religious practices. Though tribal religions often have some affinity with Hinduism, many are independent in their own rights. Table 14 provides the composition of population by religious groups and the rural-urban split of each religious group as per the expenditure survey of 2009-10. Hindus comprise 82 percent of the population, Muslims 12.8 percent, Christians 2.3 percent, Sikhs 1.7 percent, Jains 0.3 percent and Zoroastrians and others the remaining 0.3 percent.
Together, Hindus and Muslims account for almost 95 percent of India's total population. With 34 percent of the population in urban areas compared to 26 percent in the case of Hindus, Muslims are more urbanized than Hindus. Among the other communities, Jains and Zoroastrians are largely an urban phenomenon. Moreover, whereas Muslims can be found in virtually all parts of India, other smaller minority communities are geographically concentrated. Sikhs are principally in Punjab; Christians in Kerala and adjoining southern states; Zoroastrians in Maharashtra and Gujarat; and Jains in Gujarat, Rajasthan, Karnataka and Tamil Nadu. Given their small shares in the total population and geographical concentration, random sampling of households in the expenditure surveys yields less than 100 observations for minority religious communities, other than Muslims, in the vast majority of the states. Indeed, as Table 15 indicates, only 13 of the 17 largest states had a sufficiently large number of households even for Muslims to allow poverty to be reliably estimated. Each of Orissa, Haryana, Punjab and Chhattisgarh had fewer than 100 Muslim households in the survey. Thus, we attempt poverty estimates by religious groups in the states separately for Hindus and Muslims only. We do provide estimates for the catch-all "other" category, but caution that in many cases these estimates are based on fewer than 100 observations and therefore subject to large statistical errors.  Tamil Nadu  3068  2817  5885  83  271  354  169  230  399  Madhya Pradesh  2611  1662  4273  92  248  340  28  56  84  Rajasthan  2395  1205  3600  129  267  396  59  81  140  Gujarat  1584  1406  2990  130  251  381  5  48  53  Karnataka  1825  1648  3473  189  304  493  22  82  104  Orissa  2880  991  3871  39  44  83  56  20  76  Kerala  1389  1078  2467  614  423  1037  603  345  948  Assam  1749  719  2468  779  97  876  88  15  103  Jharkhand  1388  799  2187  165  94  259  205  96  301  Haryana  1311  1105  2416  51  35  86  78  40  118  Punjab  360  951  1311  30  36  66  1170  568  1738  Chhattisgarh  1458  659  2117  6  45  51  32  32  64  Total  39450 26049 65499 5161  4563  9724  2953  2207  5160 As before, we present the estimates for the statewide poverty among the religious groups at the Tendulkar line, leaving more detailed estimates for rural and urban areas and estimates at the Lakdawala lines for the appendix. Table 16 reports the estimates for Hindus, Muslims and "other" minority religion groups for years 1993-94, 2004-05 and 2009-10.  Religious groups replicate the broad pattern seen in the context of poverty by social groups. Poverty has fallen in every single state between 1993-94 and 2009-10 for Hindus as well as Muslims though the change is not always monotonic. While the level of poverty in 2009-10 is higher for Muslims than Hindus in the majority of the states, the reverse is true in Bihar, Tamil Nadu, Madhya Pradesh and Karnataka. An anomaly is the marginal increase in the poverty rate between 2004-05 and 2009-10 in Bihar for Hindus and in Gujarat for Muslims. The observation is particularly surprising since we simultaneously observe a significant decline in poverty during the same period for Muslims in Bihar and for Hindus in Gujarat 20 . Interestingly, as the appendix documents, poverty rates for both Hindus and Muslims decline in both states at Lakdawala between 2004-05 and 2009-10.

Inequality
Although the focus of this paper is on poverty, we find it useful to briefly report the evolution of inequality at the state and national levels in rural and urban areas. At the outset it is important to note that the issue of inequality is complex in part because it can be measured in numerous ways. 21 The potential list of measures is almost endless and there is no guarantee that these different measures will move in the same direction. Therefore, it is quite easy to show simultaneously that inequality has risen as well as fallen depending on the choice of measure.
In this paper, we use one measure of overall inequality based on the same expenditure survey data we have used to report the poverty measures in the previous sections: the Gini coefficient of household expenditures in rural and urban areas in the 17 states and in India as a whole using URP expenditures in 1983URP expenditures in , 1993URP expenditures in -94, 1999URP expenditures in -2000URP expenditures in , 2004URP expenditures in -05 and 2009. Table 17 and Table 18 report the Gini coefficient in rural and urban areas, respectively. As before, we arrange the states in descending order of population.
An immediate observation from Tables 17 and 18 is that with rare exceptions, rural inequality is lower than urban inequality. At the national level the Gini in 2009-10 was 0.291 in rural areas and 0.382 in urban areas. These values reflect a difference of 9 percentage points. This is not surprising. The vast majority of the villagers are small farmers or wage laborers. As a result, variation in their incomes and therefore expenditures are not large. In contrast, cities serve as home to much of the industry and formal sector services as well as to a large informal sector, which attracts migrant workers. This results in greater variation in incomes and expenditures.
The tables show no clear trend in the Gini in rural areas but a tendency for it to rise in urban areas. At the national level, the rural Gini fell between 1983 and 1999-2000, rose between 1999-2000 and 2004-05 and fell again by 2009-10 with a small net decline over the entire period. In contrast, the urban Gini has climbed up steadily. This is hardly surprising since rapid growth, which can produce increased inequality, is concentrated in urban areas. In the Indian case, a dualism of sorts exists within urban areas. Output growth has been concentrated in the formal sector while employment has been disproportionately concentrated in the informal sector. Unlike South Korea and Taiwan, China in the 1960s and 1970s and China more recently, employment in the formal sector has not grown in India due to the poor performance of labor-intensive sectors. Growth in India has concentrated in skilled-labor-and capital-intensive sectors.  The data do not support the hypothesis that high levels of poverty reflect high levels of inequality. At least in the Indian case, the two outcomes are at best unrelated and at worst negatively associated. For example, at the national level, rural inequality has remained more or less unchanged and urban inequality has risen while both rural and urban poverty have steadily and significantly declined over time.
Looking at a cross-section of the data, Kerala offers the most dramatic example. In 2009-10, it had the lowest levels of rural and urban poverty and by far the highest rural and urban Gini coefficients. At the other extreme, Bihar had the second-lowest rural Gini but the highest rural poverty ratio in 2009-10.
At a more aggregate level, the left panel in Figure 5 plots the rural Gini against the rural poverty ratio and the right panel plots the urban Gini against the urban poverty ratio. The exponential trend line has a negative slope in each case though the fit is poor. In other words, there is no evidence of a positive relationship between poverty and inequality, but there is some evidence of a negative relationship.

Concluding Remarks
In this paper, we have provided a comprehensive analysis of poverty in India along six different dimensions: time, states, rural versus urban, social groups, religious groups and poverty lines (Lakdawala and Tendulkar). To keep the exposition manageable, we have concentrated on estimates based on the Tendulkar line except when we discuss poverty at the national level. In the latter case, we report estimates in rural and urban India at both the Lakdawala and Tendulkar lines. Our detailed estimates by social and religious groups, by rural and urban areas, and by states at both the Lakdawala and Tendulkar lines are provided in the appendix.
The following are some of the key conclusions of the paper. First, poverty has declined between 1993-94 and 2009-10 along every dimension. Indeed, poverty has fallen for every social and religious group in every state in rural and urban areas separately as well as taken jointly. Estimates at the Lakdawala line, presented in the appendix, show that the decline can be observed steadily since 1983 for all social and religious groups in all 17 large states.
Second, acceleration in growth rates between 2004-05 and 2009-10 has been accompanied by acceleration in poverty reduction. Poverty rates have fallen rapidly for all major social and religious groups at the national level. This phenomenon also holds true for most states across various social and religious groups.
Third, for the first time, poverty reduction between 2004-05 and 2009-10 has been larger for the Scheduled Castes and Scheduled Tribes than the upper caste groups. Thus, the gap in poverty rates between the socially disadvantaged and upper caste groups has narrowed over time. This pattern provides clear evidence to refute the claim that reforms and growth have failed to help the socially disadvantaged or that they are leaving these groups behind. A continuation of this trend, helped along by further reforms and higher growth rates, would help eliminate the difference in poverty rates between the historically disadvantaged and the privileged.
Fourth, inter-state comparisons reveal that the states with large Scheduled Caste and Scheduled Tribe populations face a more uphill task with regards to combating poverty. The point is most forcefully brought out by a comparison of Punjab and Kerala. When we compare poverty rates in 2009-10 by social groups, the two states have very similar poverty rates. But because the poverty rates for the Scheduled Castes are higher than those for non-scheduled castes in both states and the Scheduled Castes account for a much larger proportion of its population, the aggregate poverty rate in Punjab turns out to be significantly higher.
Finally, we find that in the Indian case, there is no robust relationship between inequality and poverty. Indeed, to the extent that such a relationship exists, it suggests that more unequal states enjoy lower levels of poverty. Kerala offers the most dramatic example in this respect. It has had one of the highest Gini coefficients in rural as well as urban areas and also one of the lowest poverty ratios in both regions. In 2009-10, its Gini coefficients were by far the highest among the large states in both rural and urban areas and poverty ratios the smallest.
In our discussion on redistribution, we make references to differentiating between those who are poor and those who live in extreme poverty. In this paper, we provide comprehensive data on the distribution of the poor across various dimensions, but we do not disaggregate the poor according to the extent of their poverty. This remains important, especially with regard to debates around redistribution, and we aim to delve deeper into these and other issues in the future.