Jump to ContentJump to Main Navigation
India’s Reforms: How They Produced Inclusive Growth$

Jagdish Bhagwati and Arvind Panagariya

Print publication date: 2012

Print ISBN-13: 9780199915187

Published to Oxford Scholarship Online: May 2012

DOI: 10.1093/acprof:oso/9780199915187.001.0001

Subscriber Login

Forgotten your password?

Growth, Openness, and the Socially Disadvantaged

Chapter:
(p.186) Chapter 5 Growth, Openness, and the Socially Disadvantaged
Source:
India’s Reforms: How They Produced Inclusive Growth
Author(s):

Megha Mukim

Arvind Panagariya

Publisher:
Oxford University Press
DOI:10.1093/acprof:oso/9780199915187.003.0005

Abstract and Keywords

Critics argue that even as poverty has fallen in aggregate, it has risen among the socially disadvantage groups such as the Scheduled Castes (SC) and Scheduled Tribes (ST). The paper challenges these critics by studying the evolution of poverty within the SC, ST, and Non-Scheduled castes over the four thick expenditure surveys conducted in 1983, 1987-88, 1993-94, and 2004-5. The authors find that poverty ratio actually fell for every single group between every two successive surveys in rural and urban areas. During the twenty-one years covered by the study, poverty declined by twenty percentage points for the SC and eighteen percentage points for the ST. The authors also find the poverty levels among different social groups to decline in all major states. They find no evidence whatsoever in favor of the hypothesis that rising incomes or openness have adversely impacted poverty among any one of the groups. As in chapter 4, they find that one or more measures of openness have had a statistically significant and favorable impact on poverty levels in the SC and NS in rural and urban regions and both regions taken together. As regards the ST, they find statistically significant effect of openness on poverty in urban areas only.

Keywords:   SC, ST, poverty, openness, states, expenditure surveys, NSS

There is now broad agreement that declining levels of poverty have accompanied sustained rapid growth in India during the last three decades. Recent econometric work by Hasan, Mitra, and Ural (2006–7) and Cain, Hasan, and Mitra (2010) has also shown that openness and labor market flexibility have contributed positively to poverty alleviation. Contrary to the prior findings of Topalova (2007), these authors find no evidence that states and regions that were more exposed to trade liberalization on account of greater employment in import-competing sectors experienced a slower reduction or an increase in poverty. On the contrary, to the extent that a statistically significant relationship between poverty and trade liberalization can be found, the evidence points to greater exposure to trade leading to larger reductions in poverty.

There remains deep skepticism on the part of the nongovernmental organizations (NGOs) and journalists, however, about growth and openness having done much to alleviate poverty among the socially disadvantaged groups referred to as the Scheduled Castes (SC), Scheduled Tribes (ST), and Other Backward Castes (OBC).1 For example, a submission by the National Campaign on Dalit Human Rights to the House of Commons of the UK Parliament published on (p.187) January 14, 2011, states, “In spite of high economic growth rates the poverty rate among excluded communities in India has increased, coupled with the insecurity of livelihoods.” In the same vein, writing in the Financial Chronicle (December 29, 2010), journalist Praful Bidwai asserts, “Rising inequalities highlight what is wrong with India’s growth trajectory, driven as it is by elite consumption and sectoral imbalances, which exclude disadvantaged groups from the benefit of rising GDP, while aggravating income disparities” (emphasis added).2

Even though the concerns in these statements are couched in terms of equity, it is clear that at a deeper level they relate to the impact of growth on poverty among the disadvantaged groups. Unfortunately, the scholarly literature analyzing the impact of reforms and accelerated growth on poverty among the traditionally disadvantaged groups remains sparse. The focus of the literature devoted to studying the changes in the fortunes of the socially disadvantaged groups during the postreform era has been on intergroup inequality rather than poverty. For example, Kijima (2006) studies whether the gap between the average consumption levels of the SC and ST households, on the one hand, and Non-Scheduled (NS) households, on the other, declined between 1983 and 1999–2000 and, if so, whether this decline could be attributed to reduced discrimination. She answers the former question in the positive but the latter in the negative. Hnatkovska, Lahiri, and Paul (2010) offer a much more comprehensive analysis of intergroup inequality, asking whether the wages, education levels, and occupational structure between the SC and ST as a group and the NS have converged. They answer forcefully in the affirmative on each count and attribute it to “the rapid structural changes in the Indian economy over the past 25 years” (p. 42).3

To our knowledge, the only study that comes close to addressing the evolution of poverty as opposed to inequality in the postreform era is Sundaram and Tendulkar (2003). These authors study the (p.188) change in poverty levels of the SC, ST, and NS households, distinguished according to the means of livelihood, in rural and urban India at the national level between 1993–94 and 1999–2000.4 They find that during these years, the ST in rural areas experienced the lowest decline in poverty, while those in urban areas saw a rise in poverty. In contrast, the SC households, whether in rural or in urban areas and whether agricultural laborers or in other occupations, experienced poverty reduction matching that experienced by the rural population on average. The authors conclude (p. 5275), “This fact holds the important message that the benefits of growth have indeed been accessed by the socially and economically disadvantaged groups of India. It highlights the fact that a growth-centered strategy for poverty reduction in India can and must be followed.”

In this chapter, we undertake a comprehensive analysis of poverty among various social groups and its relationship to growth and openness. Our main results may be summarized as follows. First, whereas the distribution of the SC between rural and urban areas and across states is broadly similar to that of the general population, ST individuals are distributed differently. Based on the 61st National Sample Survey (NSS) round, conducted in 2004–5, only 8 percent of ST individuals compared with 20 percent of SC individuals live in urban areas. Likewise, the top five states ranked by their proportion of SC population mirror the five most populous states. But only one state among the top five ranked by their proportion of ST population is among the five most populous states. In other words, the distribution of the SC follows that of the general population, while that of the ST is very different from it.

Second and more importantly, nationally, every successive sample survey we analyze—that is, the 38th, 43rd, 50th, and 61st rounds, conducted in 1983, 1987–88, 1993–94, and 2004–5, respectively—shows a declining rate of poverty for every social group in every (p.189) region (rural, urban, and overall). In aggregate, the poverty rate fell by twenty percentage points for the SC and by eighteen percentage points for each of the ST and the NS between 1983 and 2004–5. Therefore, data forcefully refute any claims that higher growth in recent decades has not been associated with any benefits for disadvantaged social groups. What can be justifiably argued is that since the overall poverty rates were much higher for the ST and SC than for the NS in 1983 and that since the percentage-point reductions have been broadly equal across the groups, the proportionate differences in the poverty rates for the SC and ST on the one hand and the NS on the other have risen.

Third, our econometric analysis offers strong evidence of rising incomes leading to declining rates of poverty for the ST, NS, and all groups taken together. Our results for the SC are weaker: while the coefficient of per capita income has the correct sign, it is not statistically significant. These results are based on regressions of poverty on per capita income with appropriate control variables based on pooled data for the eighteen largest states that account for 98 percent of the national population over our four survey years.

Finally, we test for the effect of trade openness on poverty among various social groups. Our results show that for no social group did poverty fall less in states that were exposed to greater openness following trade liberalization. To the extent we find statistically significant effects, they show greater reductions in poverty in the states that were more exposed to openness. Specifically, one or more measures of protection shows a negative and statistically significant effect of openness on poverty for the SC and NS in urban, rural and urban and rural areas combined. For the ST, we find statistically significant effects in urban areas only.

This chapter is organized as follows. In the next section, we describe the size and the location of the socially disadvantaged groups. In the third section, we report the estimates of poverty at the national (p.190) level for the three major groups in various regions (rural, urban, and combined) for the four surveys that we analyze. In the fourth section, we describe poverty trends based on estimates for the top ten states by SC and ST populations. At this stage we also estimate the relationship between poverty and per capita incomes. In the fifth section, we test the relationship between openness and poverty alleviation. In the sixth section we perform some sensitivity analysis and in the final section we conclude and discuss the implications of our findings.

The Socially Disadvantaged: How Many and Where?

Demographic data on the SC and ST come from two sources: the Census Commission and the National Sample Survey Office (NSSO). The counts from these sources do not match and they also relate to different years. Because virtually all of our quantitative analysis relies on unit-level data from the NSSO, we report the indicators based predominantly on this source. But where relevant, we do point out the differences with the census data.

Though the NSSO conducts a smaller, “thin” expenditure survey each year, it conducts a larger, “thick” survey approximately every five years only. Our analysis is based on these quinquennial thick surveys beginning with the one conducted in 1983. We consider five NSSO surveys in all relating to the years 1983, 1987–88, 1993–94, 1999–2000, and 2004–5 and labeled rounds 38, 43, 50, 55, and 61, respectively.5 Of these, the expenditure data generated by the 55th round, relating to the year 1999–2000, are not directly comparable to those in other rounds due to the application of a different sample design (see Panagariya 2008, pp. 136–41, for details). Therefore, we do not consider the 55th round except for the purpose of reporting the population shares of various social groups, which are not influenced by (p.191) the difference in the sample design. The bulk of our analysis is focused on rounds 38, 43, 50, and 61.

How Many?

The first three of the latest five thick rounds identify only the SC and ST households as separate social groups. The last two rounds identify additionally the Other Backward Castes (OBC). Accordingly, in Table 5.1, we report the proportions of the SC and ST in the total population in all five rounds and those of the OBC in the last two rounds. The residual group is labeled Forward Castes (FC), with the Non-Scheduled (NS) category defining the total population minus the SC and ST or, equivalently, the OBC and the FC. Unless otherwise stated, all our indicators drawn from the NSSO expenditure surveys are based on the eighteen largest states (counting Delhi as a state) for the first three surveys and on the twenty-one largest states for the last two. The latter include the states of Chhattisgarh, Jharkhand, and Uttarakhand, carved out of Madhya Pradesh, Bihar, and Uttar Pradesh, respectively, in 2000. We exclude the six smallest northeastern states, Sikkim, and all union territories (UTs) except Delhi.6 For comparison, we also report the data from the 2001 census, which are based on all states and UTs.

Table 5.1 Shares of Various Social Groups in the Population

Survey/Census Year

SC

ST

OBC

FC

NS

Total Population (million)

1983

16.7

8.3

74.9

660.4

1987–88

16.6

9.2

74.1

648.7

1993–94

19.4

8.5

72

767.3

1999–2000

19

8.3

36.1

36.6

72.7

904.5

2004–5

19.7

8.1

41.2

30.9

72.1

968

Census 2001

16.2

8.2

75.6

1029

Source: Authors’ calculations using the unit-level data of the relevant NSS expenditure surveys and Table T00-005 from the Census of India website at http://www.censusindia.gov.in/Tables_Published/A-Series/A-Series_links/t_00_005.aspx (accessed February 9, 2011)

Table 5.1 gives rise to four observations. First, the share of the SC in the population could be anywhere between 16 and 20 percent. If we go by the census and the two earlier NSS rounds, the proportion is closer to 16 percent, but if we rely on the three latest NSS rounds, it is closer to 20 percent. We do not take a position between these two numbers, letting the reader make this choice. Second, the share of the ST is between 8 and 9 percent. Thankfully, there is strong agreement among the five surveys and between them and the census in this regard. Third, the OBC are likely to be somewhere between 36 (p.192) and 41 percent of the total population. This is a wide range, and we must await the results of the 66th NSS round and the 2011 Caste Census for some resolution of whether the true share is nearer the lower or upper limit or some figure outside this range. Finally, as a word of caution, we note that there is wide discrepancy in the total population figures generated by the NSSO and the census. It is generally recognized that while the NSSO generates good estimates of various proportions, the census produces more accurate absolute totals. Therefore, for example, when the Planning Commission calculates the absolute number of poor, it multiplies the estimated poor from NSS data by a factor equaling the census-based total population divided by the NSS-based total population.

(p.193) Where: The Rural-Urban Split

Next, we turn to the rural-urban split of shares of various social groups. For this purpose, we rely exclusively on the 2004–5 NSSO survey, which is the latest available survey that explicitly identifies the share of the OBC. Table 5.2 summarizes the information.

Table 5.2 Rural-Urban Categorization across Social Groups, 2004–5

Region

SC

ST

OBC

FC

NS

Total

SHARES IN THE TOTAL POPULATION

Rural

15.7

7.5

32.2

19.3

51.5

74.7

Urban

4

0.7

9

11.6

20.6

25.3

Rural + Urban

19.7

8.1

41.2

30.9

72.1

100

RURAL-URBAN SPLIT WITHIN EACH CASTE CATEGORY

Rural

79.8

91.9

78.1

62.4

71.4

74.7

Urban

20.2

8.1

21.9

37.6

28.6

25.3

Rural + Urban

100

100

100

100

100

100

Source: Authors’ calculations from the 61st round of the NSSO expenditure survey

Four features of this table are worthy of note. First, at .7 percent, the share of the ST residing in the urban areas in the total population of the twenty-one largest states is tiny. Seen another way, as much as 92 percent of the total ST population in the country lives in the rural areas. In this sense, any targeted programs to improve the fortunes of (p.194) the ST require focus on rural areas though care must be taken to ensure that these programs do not impede migration to urban or other rural areas.7 Second, though the OBC are more numerous than the SC, their rural-urban composition is almost the same as that of the latter. Third, together the SC, ST, and OBC account for as much as 74 percent of the rural population. This means that any policy that lowers rural poverty is almost sure to help one or more of these groups. Finally, the FC as a group is significantly more urbanized than the remaining social groups. It accounts for almost 46 percent of the urban population, even though its share in the total population is just 30.5 percent.

Where: The Statewise Distribution of Various Groups

Turning to the geographical distribution of various social groups, we rely once again on the 61st round, conducted in 2004–5, since it is the latest survey providing data on the OBC. We consider two sets of shares: the share of each state in any given social group and the share of each group in any given state. These are shown in Tables 5.3 and 5.4, respectively. For reasons outlined above, we limit ourselves to the twenty-one largest states (counting Delhi as a state) in each table. These states account for more than 98 percent of the total population of the country. The excluded entities include six northeastern states other than Assam, Sikkim and Goa, and the six Union Territories (UTs). We arrange the states in Tables 5.3 and 5.4 in the descending order of their total population shares within the twenty-one included states. Thus Uttar Pradesh, the most populous state, appears at the top of the list, and Himachal Pradesh, the least populous state, appears at the bottom.

Several observations follow from Table 5.3. First, the five largest states also have the highest shares of the SC population, although the individual rankings differ somewhat. These states—Uttar Pradesh, (p.195) (p.196) Maharashtra, West Bengal, Bihar, and Andhra Pradesh—account for 50 percent of the total population and 54 percent of the SC population within the twenty-one largest states. Going further down the list, we see that the nine largest states continue to account for the top nine shares in the SC population. The shares of these nine states in the total and SC populations are 73 and 77 percent, respectively. It is only when we get to the tenth state that the pattern is broken, since Punjab, which is the fifteenth largest state, happens to have the tenth largest SC population. As we will see below, the SC account for an unusually large proportion of the total population in Punjab.

Table 5.3 Shares of States in Social Groups

State*

SC

ST

OBC

FC

NS

All

Uttar Pradesh

20

1

21.8

13

18.1

17

Maharashtra

7.6

11

7.2

13.6

9.9

9.5

West Bengal

11.1

6.5

1.3

15.9

7.5

8.2

Bihar

8.4

0.5

11

4.4

8.2

7.6

Andhra Pradesh

7

6.5

8.6

6.8

7.8

7.5

Madhya Pradesh

5.5

16.2

5.9

4.4

5.3

6.2

Tamil Nadu

6.5

0.4

10.1

0.9

6.2

5.8

Rajasthan

6.1

9.2

6.2

4

5.2

5.7

Karnataka

4.6

4.1

4.8

6

5.3

5.1

Gujarat

2.6

8.8

4.6

5.6

5.1

4.9

Orissa

3.3

11

3.6

2.7

3.2

3.8

Kerala

1.7

0.6

4.6

2.9

3.9

3.2

Assam

1.3

5.7

1.1

4.6

2.6

2.6

Jharkhand

1.6

8.2

2.7

1.3

2.1

2.5

Punjab

4.4

0.1

1.2

3.3

2.1

2.4

Haryana

2.9

0.1

1.6

3.2

2.3

2.2

Chhattisgarh

1.6

8.8

2.4

0.7

1.6

2.2

Delhi

1.6

0.2

0.4

2.6

1.3

1.3

Uttarakhand

0.9

0.5

0.4

1.6

0.9

0.9

Jammu and Kashmir

0.5

0.1

0.2

1.7

0.8

0.7

Himachal Pradesh

0.8

0.4

0.2

1.1

0.6

0.6

All twenty-one states

190.9

78.9

398.9

299.3

698.2

968

(Population in million)

*States are listed in the descending order of size according to total population.

Source: Authors’ calculations from the 61st round of the NSSO expenditure survey

Second, the distribution of the ST differs significantly across states. The five largest states, accounting for 50 percent of the total population, account for only 25.5 percent of the ST population in the twenty-one states. The state with the highest share in the ST population of the (p.197) (p.198) twenty-one states, Madhya Pradesh, is not among the five largest states by total population. Nevertheless, the degree of concentration of the ST is almost as high across states as of the SC population. The top five and top nine ST states account for 56 and 86 percent of the ST population in the twenty-one states, respectively. The top five ST states are Madhya Pradesh, Maharashtra, Orissa, Rajasthan, and Gujarat, in that order, with Chhattisgarh a close sixth. Only Maharashtra appears on both the top five SC and ST lists. Remarkably, none of the seven northeastern states, traditionally identified as those with high concentrations of the ST populations, appear on this list. These states do have very high proportions of the ST in their overall populations but being small in size, they house only a small proportion of the country’s total (p.199) ST population. Assam, the largest northeastern state, accounts for only 6 percent of the ST population in the largest twenty-one states.

Third, the distribution of the OBC across states follows more closely the distribution of the SC. Nine states on the top ten SC and OBC lists are common. The top five and top nine OBC states account for 59 and 80 percent of the total OBC population in the twenty-one states, respectively. This close correspondence between distributions of the SC and OBC perhaps has a bearing on the political economy of reservations in jobs and education: living side by side with the SC populations, which enjoyed the benefits of reservations since the adoption of the Constitution in 1950, the OBC populations may have become sensitized to these advantages and perhaps actively sought similar reservations for themselves.

Finally, the distribution of the FC resembles to some degree that of the SC and OBC, though with greater concentration at the top and more even distribution over the remaining states. The top three FC states, which are also the three most populous states (Uttar Pradesh, Maharashtra, and West Bengal), account for 43 percent of the FC population in the twenty-one states, in comparison to 39 percent for the SC, 40 percent for the OBC, and 28 percent for the ST. There is a steep decline as we move from the third-largest FC state (Uttar Pradesh) to the fourth-largest (Andhra Pradesh): from 13 to 7 percent.

Some insight into the dispersion of various social groups across states can be gained with the help of the Theil index of inequality. This index belongs to the family of generalized entropy inequality measures. The value varies between 0 and ∞, with zero representing an equal distribution and higher values representing higher levels of inequality. The index allows us to identify separately the contribution of various entities to the overall concentration.

Using the demographic data from the 61st NSS round, conducted in 2004–5, the value of the index for the total population is 3.63, (p.200) which provides a benchmark for the distribution of the population across different states in India. The values of the index for the SC and NS turn out to be close to this value: 4.46 and 3.85, respectively. But consistent with the observations above, the value of the index for the ST turns out to be much higher at 6.3.

The Theil index allows us to calculate the contributions of individual states to the final value of the index—states that host a large proportion of the population of any given social group make a larger contribution to the index. The maps in Figure 5.1 provide a visual depiction of each state’s contribution to the Theil index by social group. While the contribution of states to the index for the NS is very similar to that of the general population (“All”), those of the SC and the ST diverge from the latter. The ST are more heavily concentrated in states such as Madhya Pradesh, Maharashtra, and Orissa, while the SC are more concentrated in Uttar Pradesh, West Bengal, and Bihar. In other words, even after controlling for the average dispersion across states, the ST individuals are strongly concentrated, and the SC individuals moderately concentrated, across states.

                      Growth, Openness, and the Socially Disadvantaged

Figure 5.1. State Contributions to the Theil Index (All and NS)

                      Growth, Openness, and the Socially Disadvantaged

Figure 5.1 (continued). State Contributions to the Theil Index (SC and ST)

Where: The Groupwise Distribution of Population within States

Table 5.4 shows the shares of various social groups in the population of each of the twenty-one largest states. Several observations follow.

Table 5.4 Shares of Groups within Each State

State*

SC

ST

OBC

FC

NS

Population (millions)

Uttar Pradesh

23.1

0.5

52.8

23.6

76.4

165

Maharashtra

15.8

9.4

30.9

43.9

74.8

92.3

West Bengal

26.8

6.5

6.5

60.3

66.7

78.9

Bihar

21.9

0.6

59.7

17.8

77.5

73.6

Andhra Pradesh

18.3

7

46.9

27.8

74.7

72.9

Madhya Pradesh

17.4

21.2

39.4

22

61.4

60.1

Tamil Nadu

22.2

0.6

72.2

5

77.2

56.1

Rajasthan

20.9

13.1

44.6

21.4

66

55.3

Karnataka

17.9

6.6

39

36.4

75.4

49.3

Gujarat

10.5

14.8

39.2

35.5

74.8

47.2

Orissa

17

23.4

38.2

21.4

59.6

37.2

Kerala

10.5

1.6

60.2

27.7

87.9

30.8

Assam

9.9

17.8

17.6

54.6

72.2

25.2

Jharkhand

12.9

26.6

45

15.5

60.5

24.3

Punjab

35.9

0.4

20.4

43.3

63.6

23.2

Haryana

25.5

0.3

30.3

44

74.2

21.6

Chhattisgarh

14.2

32.4

44.1

9.3

53.4

21.5

Delhi

24.6

1.6

11.4

62.5

73.9

12.4

Uttarakhand

21.4

4.8

18

55.8

73.8

8.3

Jammu and Kashmir

12.7

0.6

12.5

74.2

86.7

6.8

Himachal Pradesh

26.2

4.9

14.9

54.1

68.9

6.1

All twenty-one states

19.7

8.1

41.2

30.9

72.1

968

*States are listed in the descending order of size according to total population.

Source: Authors’ calculations from the 61st round of the NSSO expenditure survey

First, Chhattisgarh, Orissa, Jharkhand, and Madhya Pradesh, in that order, stand out in terms of having large numbers of both the SC and ST in their populations. The SC and ST also account for a significant proportion of the population in Rajasthan. Second, in the remaining states, it is either the SC or the ST that has a major presence, but not both. The SC account for 20 percent or more of the population in as many as ten states. Finally, and somewhat surprisingly, the OBC constitute the largest single social group in eleven out (p.201) (p.202) (p.203) of the fourteen most populous states. The FC constitute the largest single group in nine out of the twenty-one states, but only two among them, Maharashtra and West Bengal, make it to the list of the twelve largest states.

Counting the Poor by Social Groups

We now turn to counting the poor by social group. As previously noted, Meenakshi, Ray, and Gupta (2000) have calculated the poverty rates for the SC and ST for 1993–94 at both national and state levels. Sundaram and Tendulkar (2003) have done the same at the national level, on a comparable basis, for 1993–94 and 1999–2000, showing declining poverty levels across the surveys. In this chapter, we offer estimates for the SC, ST, and NS (Non-Scheduled, consisting of the OBC and the FC) for 1983, 1987–88, 1993–94, and 2004–5 and for the OBC and FC separately for the year 2004–5. We choose to skip the year 1999–2000 because of noncomparability of the survey design that year to the surveys in other years. For comparability with the overall official poverty estimates published by the Planning Commission, we base all our estimates on the official poverty lines.

Poverty Rates across Groups Nationally and Statewise

We first present the poverty rates in 2004–5 across various social groups in rural and urban regions aggregated over the twenty-one largest states (counting Delhi as a state). Several observations follow from Table 5.5 with respect to the latest poverty picture. First, the poverty rates are the highest among the ST followed by the SC, the OBC, and then the FC in that order. The only exception is the urban poverty rate for the ST, which is a hair’s breadth below the corresponding rate (p.204) for the SC. But even this exception has limited relevance, since only 8 percent of the ST population lives in the urban areas. Second, the SC and ST poverty rates are an order of magnitude higher than either the average for all groups or the OBC and FC. Finally, for both the SC and OBC, the urban poverty rates are higher than the corresponding rural poverty rates, and for both groups the urban and rural poverty rates are substantially higher than the corresponding rates for the FC. It is possible that with the existence of antipoverty programs that concentrate mainly in rural areas and with rural-to-urban migration predicted to accelerate in coming years, urban poverty rates for the SC and OBC would remain higher than the corresponding rural poverty rates in the short to medium run.

Table 5.5 Poverty Rates by Social Groups in the Twenty-One Largest States, 2004–5

Region

SC

ST

OBC

FC

NS

All Groups

Rural

37.2

47

25.9

17.5

22.8

28.2

Urban

41.1

39

31.3

16.2

22.8

26.1

All (rural + urban)

38

46.3

27.1

17

22.8

27.7

Source: Authors’ calculations using the unit-level expenditure data from the 61st round

We also find it useful to present the poverty picture by states in 2004–5 using maps. In Figure 5.2, we show the poverty rates for the general population and for the NS, SC, and ST social groups across states. Before describing the maps, we note one important limitation of some of our poverty estimates for the ST at the state level. In several states, the NSS survey ends up with only a small number of (p.205) (p.206) (p.207) ST households. Five extreme cases are Goa, Haryana, Delhi, Punjab, and Jammu and Kashmir (J&K), which ended up with just 2, 11, 16, 18, and 20 ST households, respectively, in the 2004–5 survey. Poverty estimates based on such small number of households are unlikely to be precise. Unfortunately, there is no way around this, at least for the purposes of comparing the estimates across states and within states over time. Therefore, for now, we choose to report the estimates for all twenty-one states, while recognizing that the estimates are less precise in some cases than in others. Later, when we carry out regressions using state-level poverty estimates, we will conduct robustness tests by excluding states for which the number of ST household in the sample happens to be small.

With this caveat, we may now proceed to identify two features of the maps presented in Figure 5.2. First, the upper limits of poverty rate are higher for the SC and ST groups compared to those for the general population and the NS. And second, the overall (i.e., rural plus urban) rate of poverty is highest in central and northeastern states, with a few exceptions. The results for the general population and the NS are similar—poverty rates range between 5 and 47 percent for the general population and 5 and 36 percent for the NS, with states such as Orissa, Bihar, Jharkhand, Chhattisgarh, Uttarakhand, and Madhya Pradesh exhibiting the highest rates of poverty. On the other hand, the upper limit on poverty is 64 percent for the SC and 75 percent for the ST. However, the highest rates of poverty for the SC and ST are associated with the same states as the overall and NS rates. For the SC, states like Bihar, Uttarakhand, Jharkhand, and Orissa, and for the ST, states like Orissa, Madhya Pradesh, Bihar, and Maharashtra suffer from the highest rates of poverty. Thus, although the rates of poverty are higher for the SC and ST, there is substantial geographical overlap when considering high poverty rates among these groups and the NS.

                      Growth, Openness, and the Socially Disadvantaged

Figure 5.2. Poverty Rates by State in 2004–5 (All and NS)

                      Growth, Openness, and the Socially Disadvantaged

Figure 5.2 (continued). Poverty Rates by State in 2004–5 (SC and ST)

(p.208) Evolution of National Poverty Rates by Social Groups over Time

In the previous section, we provided snapshots of relative poverty rates across various groups at the national and state levels in 2004–5. Next, we turn to the evolution of poverty rates for various social groups over time at the national level. Since the OBC are identified only in 2004–5, for the purpose of comparison over time, the social groups must be limited to the SC, ST, and NS. Table 5.6 reports the relevant rate of poverty in rural, urban, and rural-plus-urban regions for various social groups for the years 1983, 1987–88, 1993–94, and 2004–5.

Table 5.6 Evolution of Poverty Rates by Social Groups

Survey Year

SC

ST

OBC

FC

NS

All Groups

RURAL

1983

59

64.9

41

46.6

1987–88

50.1

57.8

32.8

38.7

1993–94

48.4

51.6

31.3

37

2004–5

37.2

47

25.9

17.5

22.8

28.2

URBAN

1983

56.2

58.3

40.1

42.5

1987–88

54.6

56.2

36.6

39.4

1993–94

51.2

46.6

29.6

33.1

2004–5

41.1

39

31.3

16.2

22.8

26.1

RURAL + URBAN

1983

58.5

64.4

40.7

45.7

1987–88

50.8

57.6

33.8

38.8

1993–94

48.9

51.2

30.8

36

2004–5

38

46.3

27.1

17

22.8

27.7

Source: Authors’ calculations using the unit-level data the NSSO expenditure surveys

Perhaps the most remarkable feature of Table 5.6 is the declining rates of poverty over time for every single group in every region (rural, urban, and overall) between every pair of surveys. The results thoroughly counter any claims that the accelerated growth occurring since the early 1980s has failed to help disadvantaged groups. Most pointedly, poverty rates have declined significantly even for the ST, who are often said to be outside the mainstream of the economy. During the twenty years covered by the surveys, poverty declined by twenty percentage points for the SC and eighteen percentage points for both the ST and NS. Critics would no doubt like to argue that given the higher initial rates of poverty for the SC and ST, these reductions imply that the ratio of poverty rates for the SC and ST to that for the NS has gone up; however, the estimates in Table 5.6 refute the claim that growth has bypassed socially disadvantaged groups.

Cross-state Analysis: Poverty and Per capita Income

We now move on to a more disaggregated analysis over time, focusing on individual states. We focus on two main issues: (1) How have SC and ST poverty rates evolved between 1983 and 2004–5 in states (p.209) (p.210) where they are highly concentrated? and (2) Are increases in average per capita income associated with declining rates of poverty at the level of the state?

Turning to the first question, we show that the trend of declining poverty rates for each group observed at the national level is broadly reproduced at the level of the states. Figure 5.3 shows the poverty rates for the SC for the ten largest states based on the statewise SC populations in 2004–5. States are arranged in declining order of SC population from left to right using data from the 61st round, conducted in 2004–5. Therefore, Uttar Pradesh has the largest SC population, West Bengal the second largest, and so on. For comparison, we also include the overall SC poverty rate in the eighteen largest states taken together. Note that we aggregate the three new states (Chhattisgarh, Jharkhand, and Uttarakhand) with their respective mother states (Madhya Pradesh, Bihar, and Uttar Pradesh).

                      Growth, Openness, and the Socially Disadvantaged

Figure 5.3. SC Poverty Rate from 1983 to 2004–5 in the Top Ten States by the SC Population

Three observations follow from Figure 5.3. First, some of the states with large SC populations and above-average SC poverty rates are among some of the poorest states in India. Uttar Pradesh, Bihar, and Madhya Pradesh were the bottom three states by per capita income in 2004–5 and are also among the top five states in terms of SC population. They also have poverty rates well in excess of the average poverty rate of the eighteen largest states by population. These three states also experienced growth rates well below the national average between 1983 and 2004–5.

Second, comparing the rates between 1983 and 2004–5, poverty declined in every one of the ten states by at least ten percentage points. In some states, the decline was impressive. For instance, it fell from 70 to below 30 percent in West Bengal and from almost 70 to a little above 30 percent in Tamil Nadu between 1983 and 2004–5. By the same token, poverty reduction in some states has been extremely slow. In Bihar, the poverty rate only fell seven percentage points, from 70 percent in 1987 to 63 percent in 2004–5.

(p.211) (p.212) Finally, going by the sample surveys, the reduction in poverty has been monotonic in most but not all states. While progress between 1987–88 and 1993–94 was limited—there were even reversals in some states—visible reductions in poverty took place in each of these ten states between 1993–94 and 2004–5. This is especially interesting considering that 1991 was the year of the balance of payments crisis, and it took some years for the country to return to the 6 percent growth rate experienced between 1993–94 and 2004–5.

Next, we turn to the evolution of poverty rates for the ST by state. As in the case of the SC, we show the poverty rates associated with the four surveys in the top ten states according to the ST population in 2004–5 in Figure 5.4. The states are arranged in declining order of the ST population, from left to right. We note that the number of ST households underlying each estimate shown in the figure is well above two hundred. As such, the estimates shown are reasonably precise.

                      Growth, Openness, and the Socially Disadvantaged

Figure 5.4. ST Poverty Rates in the Top Ten States by the ST Population. 1983 to 2004–5

The pattern is similar to that for the SC, with one important difference. As in the case of the SC, there is a significant decline in poverty rates for all states between 1983 and 2004–5. But in contrast to the SC, there are several states in which ST poverty rates have seen a marginal increase between 1993–94 and 2004–5. This particular fact is contrary to the hypothesis of declining poverty rates with increasing per capita incomes.

Using poverty rates and per capita-income data, we can study more directly the relationship between poverty and growth for various social groups. For this purpose, we first present a set of graphs showing declining poverty rates for each group statewide with rising per capita incomes using cross-state data for 2004–5. We obtain similar graphs when we disaggregate the data by social group for rural and urban areas in 2004–5 and for 1983, 1987–88, and 1993–94 in rural, urban, and rural-plus-urban areas. We do not include these graphs here, to economize on space, but they are available on request. (p.213) (p.214) Following the graphs, we present some econometric results relating poverty rates to per capita incomes for each social group.

Figures 5.55.8 show the groupwise scatter plots of poverty rates against per capita incomes and the associated best-fit lines. In each graph, we compare the relationship between poverty level and per capita income for an individual group such as the SC or ST to that for the entire population referred to as “All.” In all cases, we combine Chhattisgarh, Jharkhand, and Uttarakhand with their respective mother states. This brings the total number of states in the sample down to eighteen. In Figures 5.5, 5.7, and 5.8, we use the observations from all of the eighteen states. But in Figure 5.6, which depicts poverty among the ST and that among all groups combined, we drop four states—Punjab, Haryana, Delhi, and J&K—from the sample because of the relatively small number of ST households picked up by the survey in these states.

Interestingly, the best-fit line between poverty rates and per capita incomes is uniformly negatively sloped in every single case. That is to say, poverty rates decline as per capita incomes rise for every social group. As we have already noted, poverty levels for the disadvantaged groups are generally higher than those for the general population.

                      Growth, Openness, and the Socially Disadvantaged

Figure 5.5. Poverty Rate and per Capita Income: SC and All Groups, 2004–5

                      Growth, Openness, and the Socially Disadvantaged

Figure 5.6. Poverty Rates and per Capita Income: ST and All Groups (with Haryana, Punjab, Delhi, and J&K removed from the sample due to small number of households), 2004–5

                      Growth, Openness, and the Socially Disadvantaged

Figure 5.7. Poverty Rates and per Capita Income: OBC and All Groups, 2004–5

                      Growth, Openness, and the Socially Disadvantaged

Figure 5.8. Poverty Rates and per Capita Income: NS and All Groups, 2004–5

We next turn to a more formal verification of the hypothesis that rising per capita incomes leading to declining poverty among various social groups. We carry out the regressions by pooling the data on poverty rates and per capita incomes for the four years for which we have survey data. For consistency, we continue to aggregate the three states (Chhattisgarh, Jharkhand, and Uttarakhand) that were created in 2000 and for which data are reported separately in the 61st round with their respective mother states (Madhya Pradesh, Bihar, and Uttar Pradesh).

We run an ordinary least squares (OLS) regression of the following form: (p.215)

(1) ln p o v i t k = α + β ln P C I i t + δ i + μ t + ε i k t

Here, the dependent variable is the poverty ratio for social group k residing in state i in year t, and the main explanatory variable is per capita income that varies by state (δi) and year (μt). Using the data for eighteen states for four years (1983, 1987–88, 1993–94, and 2004–5), we estimate equation (1) for the SC, ST, and NS. We measure per capita income by per capita Net State Domestic Product (NSDP). At this stage, we continue to use observations for all social groups, including the ST for all eighteen states. Later, we will consider the implications of dropping the ST observations associated with unacceptably imprecise estimates of poverty due to small number of households appearing in certain states.

We report the results from four sets of regressions in Table 5.7: without any fixed effects, with year-fixed effects, with state-fixed effects, and with both year- and state-fixed effects. In the first three sets of regressions, eight of the nine coefficients are statistically significant at the 99 percent level and show that poverty is indeed negatively related to per capita income. Even in the remaining case, the sign of the coefficient supports a negative relationship, but owing to a high standard error, it is not statistically significant. The magnitudes of the coefficients are all within the plausible range: a 100 percent increase in income leads to an eleven- to twenty-seven-percentage-point reduction in poverty.

Table 5.7 Per Capita Income and Poverty with and without Year- and State-Fixed Effects

Item

SC

ST

NS

(1) NO FIXED EFFECTS

Coefficient

–16.0321**

–25.0258***

–19.6469***

Standard Error

6.0

5.92

3.13

R2

0.24

0.31

0.51

(2) STATE-FIXED EFFECTS ONLY

Coefficient

–23.2982***

–17.8310***

–19.2071***

Standard Error

3.7

4.89

2.37

R2

0.85

0.78

0.9

(3) YEAR-FIXED EFFECTS ONLY

Coefficient

–11.18

–27.0780***

–18.9509***

Standard Error

8.46

8.71

4.57

R2

0.3

0.32

0.53

(4) STATE- AND YEAR-FIXED EFFECTS

Coefficient

–1.67

4.72

–4.71

Standard Error

10.14

15.35

6.01

R2

0.89

0.8

0.93

*** p 〈 0.01, ** p 〈 0.05 and * p 〈 0.1

But in the last case, where we allow for both state- and year-fixed effects, the results change dramatically. The coefficient for none of the social groups is significant any longer, the coefficient for the ST changes sign, and the remaining two coefficients see a drastic fall in magnitude. At first blush, this might appear to be a devastating blow to the hypothesis of a causal effect running from per capita income (p.216) (p.217) (p.218) (p.219) (p.220) (p.221) to poverty. For example, there may be something specific to each state and each year other than the change in per capita income that accounts for the observed patterns in poverty as well as per capita income.

There remains the possibility, however, that with four years and eighteen states, the year- and state-fixed effects soak up all the variation in income, leaving the latter with no additional variation. We investigate this possibility by regressing per capita income on state and year dummies. The reported R2 is .958, implying that state and year dummies account for 95.8 percent of the variation in per capita income. It is small wonder, then, that the inclusion of both sets of fixed effects leaves little extra variation in per capita income. By regressing per capita income separately on state and year dummies, we find that the former by themselves account for 63.3 percent of the variation in per capita income and the latter for 32.5 percent.

Based on these findings, we proceed to control for state-specific characteristics directly rather than through the introduction of dummy variables. We continue to use year-fixed effects to control for time-varying unobserved characteristics. We identify five state-level characteristics: lagging versus leading states, landlocked versus coastal states, the proportion of the SC and the ST in the state population in the first year of observation (1983), and the proportion of all poor in the state population in 1973–74.8 Each of these variables varies across states and may impact poverty independent of per capita income. Thus, we modify the regression equation as below

(2) ln p o v i t k = α + β ln P C I i t + γ C o n t r o l s i t k + μ t + ε i k t

Our results are shown in Table 5.8, where we also include a specification for poverty for all groups (i.e., the general population) in the last column. In all four cases, the coefficient of per capita income is (p.222) (p.223) negative. In the case of ST, NS, and all groups taken together, it is also statistically significant at the 95 percent or higher level.9 It is only in the case of the SC that the coefficient remains insignificant at the 90 percent or higher level of significance. The three statistically significant estimates are within the plausible range: a 100 percent increase in per capita income is associated with a poverty reduction of 13.8 percent for the ST, 11.2 percent for the NS, and 9.5 percent for the general population.

Table 5.8 Per Capita Income and Poverty with Controls for State-Level Characteristics

Variable

SC

ST

NS

ALL

Ln PCI

–4.49

–13.7771**

–11.2362***

–9.5038***

[3.841]

[5.791]

[2.784]

[2.894]

Year 1987–88

–6.9921**

–3.5

–5.6521***

–5.9004***

[2.910]

[4.387]

[2.109]

[2.192]

Year 1993–94

–8.5048***

0.6

–5.5528**

–5.8985**

[3.072]

[4.631]

[2.227]

[2.315]

Year 2004–5

–17.7032***

–6.43

–9.1933***

–10.7609***

[3.921]

[5.911]

[2.842]

[2.954]

Lagging state

–7.2551*

10.1631*

0.25

0.29

[3.977]

[5.996]

[2.883]

[2.997]

Landlocked

8.0156**

–18.2783***

–0.76

–0.36

[3.333]

[5.025]

[2.416]

[2.511]

SC initial

0.5007**

0.6518**

0.15

0.4101***

[0.204]

[0.308]

[0.148]

[0.154]

ST initial

0.02

0.8768***

0.08

0.3233***

[0.159]

[0.239]

[0.115]

[0.120]

Poverty 1973–74

1.2598***

0.28

0.6518***

0.7325***

[0.152]

[0.229]

[0.110]

[0.114]

Constant

13.02

123.4963***

83.4670***

64.6426***

[29.625]

[44.663]

[21.472]

[22.321]

Observations

72

72

72

72

R2

0.74

0.68

0.8

0.81

Standard errors in square brackets

*** p 〈 0.01, ** p 〈 0.05, * p 〈 0.1

These results support the hypothesis that an increase in overall state-level per capita incomes leads to a reduction in the rates of poverty for the ST and NS. In the case of SC poverty there is no evidence that an increase in per capita income leads to any harm, but we would require additional data to claim with statistical confidence that an increase in per capita income has a positive impact. We remind readers that poverty estimates for the ST in some of the states included in our sample may be unacceptably imprecise due to the small number of ST households observed. We will return to this qualification later in a separate section devoted to robustness check.

(p.224) Poverty and Openness

Recent work by Hasan, Mitra, and Ural (2006–7) and Cain, Hasan, and Mitra (2010) has shown that there is a negative relationship between openness and poverty rates for all groups taken together in rural, urban, and rural-plus-urban regions. Their results reverse those obtained earlier by Topalova (2007) and reiterated in Topalova (2010). Both sets of papers exploit the variation in the degree of openness generated by India’s trade liberalization in the 1990s across different administrative units of the country. A key difference between the approaches taken by them is with regard to the measurement of openness. Topalova treats nontraded sectors as the same as freely traded import-competing sectors, while Hasan et al. treat them (correctly, in our view) as nontraded. Naturally, the differences in the assumptions lead to differences in the numerical measures of openness employed in the two sets of papers. Hasan et al. and Topalova also differ with respect to the unit of analysis: whereas the former carry out their analysis at the level of the state and region as defined by the NSSO, the latter does so at the level of the district. While district-level analysis has the advantage of higher degrees of freedom in regressions, it also has the disadvantage of reducing the number of observations on which the poverty estimate itself is based. Hasan et al. discuss other problems with district-level analysis.

A common concern expressed truculently in Indian policy circles is that even if openness helps the population overall, it is detrimental to the interests of socially disadvantaged groups. We have already shown in the previous section the impact of growth on poverty among the socially disadvantaged groups is hardly different from that on remaining groups. While openness may work partially through growth, in the present section we consider its impact on poverty more directly.

(p.225) Given that the sample size of the survey becomes much smaller when we restrict the counts to socially disadvantaged groups, we run a serious risk of large measurement errors in the estimation of poverty ratios at the level of the district. For this and other reasons discussed in Hasan et al. (2007) and Cain et al. (2010), we choose to carry out our analysis at the level of the state. This has the added advantage that it allows us to use their measures of protection. They construct state-level measures of trade protection for three regions—rural, urban, and rural plus urban. They weight industry-level tariff rates and nontariff barrier (NTB) coverage rates for two-digit agricultural, mining, and manufacturing industries by sector-specific employment shares in each state using the following formulas:10

T a r i f f i t j = k γ i k , 1993 j * I n d _ T a r i f f k t
N T B i t j = k γ i k , 1993 j * I n d _ N T B k t

Here γ i k , 1993 j is the employment share of industry k in region j (j = rural, urban, and rural plus urban) of state i derived from the 1993–94 employment-unemployment survey.11 I n d _ T a r i f f k t and I n d _ N T B k t represent tariff rates and nontariff coverage rates in industry k in year t where industries are measured at a two-digit classification. The employment share weights are defined such that they sum to unity. Stated simply, k γ i k , 1993 j = 1 where k represents tradable two-digit industries (comprising agricultural, mining, and manufacturing industries). Nontradable industries are excluded from the calculations.

Because tariff rates and nontariff barriers are highly correlated, they cannot be used simultaneously in the regressions. Therefore, we use the two measures separately and in succession. In addition, following Cain et al. (2010), we use a third measure, which combines these tariff and nontariff measures into a single measure using (p.226) principal component analysis. Principal component analysis is commonly deployed to collapse the vector of correlated variables into a smaller set of variables containing much of the variation in the data. In the present case, the first principal component contains approximately 90 percent of the variation in the protection data for all industry groups.

The basic regression equation we estimate is

y i j t k = α + β 1 p r o t e c t i o n i j t 1 + β 2 Z i t + δ i + μ t + ε i t

In this equation, the dependent variable y k i j t is the logarithm of poverty for social group k in sector j (urban, rural, and rural-plus-urban) in state i and year t. The principal variable of interest, protectionijt−1, is one of the three measures of trade protection lagged once: nominal rate of protection (NRP), nontariff barriers (NTB), and the first principal component (FPC) of NRP and NTB.12 Variable Zit denotes a time-varying state-level control variable, which we choose to be per capita development expenditures. Vectors δi and μt represent state-fixed and year-fixed effects, respectively. Finally, εit is an error term and is assumed to satisfy the usual properties.

We note that all regression results we report in this section control for year- and state-fixed effects. Our first step is to reproduce the basic results of Cain et al. (2010), which relate to all groups taken together in rural, urban, and rural-plus-urban regions. Because the data on protection at the level of the states are available only from 1986 and for fifteen states, our analysis is based on forty-five observations relating to the 43rd, 50th, and 61st NSSO rounds, conducted in 1987–88, 1993–94, and 2003–4, respectively. While we have data on poverty rates for the 38th round, conducted in 1983, the protection series does not go that far back. Likewise, while we have protection data to include the 55th round, conducted in 1999–2000, we do not have the poverty levels by social groups on a comparable basis for this year.

(p.227) Our results for all social groups combined closely correspond to those of Cain et al. (2010), although we do not include data from the 55th NSS round (1999–2000). For instance, they find that poverty declines by .57 percent for every percentage-point reduction in the weighted tariff rate (NRP). As shown in Table 5.9, the corresponding figure in our analysis is .67 percent. Our other results are similar; we describe these briefly without comparing them to those in Cain et al. (2010). Thus, according to our results, a 1 percent reduction in the NTB coverage ratio is associated with a 2.68 percent fall in the headcount ratio. The analysis for urban and rural regions shows that trade liberalization has affected them differentially. A 1 percent reduction in weighted tariff measures lowers poverty in the urban areas by 1.96 percent, but its effect on rural poverty, while having the hypothesized sign, is statistically insignificant. On the other hand, a fall in nontariff barriers in rural areas is associated with a whopping 3.7 percent reduction in poverty but their effect on urban areas is small and statistically insignificant.

Table 5.9 Poverty and Trade Openness (All Social Groups)

Combined

Urban

Rural

NRP

NTB

FPC

NRP

NTB

FPC

NRP

NTB

FPC

VARIABLES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Protection

.0067**

.0268**

.3324**

.0191**

0.02

.9627**

0.01

.0368**

.3013**

[.003]

[.012]

[.120]

[.007]

[.019]

[.361]

[.004]

[.014]

[.143]

Dev. Exp. Pc

0.14

0.2

0.13

.78807*

.7499*

.7209*

0.24

0.3

0.22

[.235]

[.230]

[.223]

[.395]

[.437]

[.393]

[.299]

[.274]

[.286]

Constant

1.64

–0.85

1.72

–3.91

–3.13

–2.84

0.77

–2.83

0.88

[1.568]

[1.959]

[1.487]

[2.798]

[3.295]

[2.668]

[1.983]

[2.316]

[1.901]

Observations

45

45

45

45

45

45

45

45

45

R 2

0.95

0.96

0.96

0.93

0.92

0.93

0.94

0.95

0.94

Standard errors in square brackets

*** p 〈 0.01, ** p 〈 0.05, * p 〈 0.1

An interesting point that emerges from our analysis of openness and poverty reduction for individual social groups, worthy of note at the outset, is that while the results for the SC closely track those for the NS group, those for the ST turn out to be much weaker. Tables 5.105.12 report our results for the SC, ST, and NS, respectively. Table 5.10 shows that a 1 percent reduction in nontariff barriers is associated with a 2.6 percent reduction in poverty in rural and urban regions combined and a 3.1 percent reduction when we consider only rural areas. Consistent with the results for all social groups, the effect of weighted tariff reductions is positive and significant (2.4 percent) for urban areas and we find a similar result for nontariff barriers in rural areas. These results are in line with the high correlation (0.7126) between the share of the general population and that of the SC across states.

Table 5.10 Poverty and Trade Openness (Scheduled Castes)

Combined

Urban

Rural

NRP

NTB

FPC

NRP

NTB

FPC

NRP

NTB

FPC

VARIABLES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Protection

0

.0258**

.2359*

.0242***

0.02

1.213***

0

.0300**

0.18

[.004]

[.012]

[.135]

[.007]

[.019]

[.347]

[.004]

[.013]

[.132]

Dev. Exp. Pc

–0.09

–0.05

–0.1

1.0103**

.9625**

.9257**

–0.23

–0.2

–0.25

[.251]

[.236]

[.243]

[.374]

[.441]

[.370]

[.267]

[.242]

[.261]

Constant

3.6533**

1.17

3.690**

–5.88**

–4.88

–4.528*

4.5747**

1.61

4.6309

[1.676]

[2.012]

[1.621]

[2.654]

[3.326]

[2.517]

[1.775]

[2.048]

[1.731]

Observations

45

45

45

45

45

45

45

45

45

R 2

0.92

0.93

0.92

0.91

0.88

0.91

0.92

0.93

0.92

Standard errors in square brackets

*** p 〈 0.01, ** p 〈 0.05, * p 〈 0.1

Table 5.11 Poverty and Trade Openness (Scheduled Tribes)

Combined

Urban

Rural

NRP

NTB

FPC

NRP

NTB

FPC

NRP

NTB

FPC

VARIABLES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Protection

0

–0.01

–0.11

.040***

0.02

1.497**

0

0

0.02

[.008]

[.031]

[.328]

[.013]

[.046]

[.631]

[.010]

[.040]

[.391]

Dev. Exp. Pc

0.75

0.73

0.75

–0.23

–0.28

–0.37

0.85

0.86

0.85

[.607]

[.600]

[.605]

[.627]

[.769]

[.676]

[.790]

[.783]

[.789]

Constant

–2.14

–1.77

–2.47

4.56

5.62

9.7702*

–3.02

–3.06

–2.96

[4.449]

[4.467]

[4.590]

[4.621]

[5.581]

[5.304]

[5.792]

[5.823]

[5.992]

Observations

44

44

44

41

41

41

44

44

44

R 2

0.72

0.73

0.73

0.91

0.87

0.9

0.64

0.64

0.64

Standard errors in square brackets

*** p 〈 0.01, ** p 〈 0.05, * p 〈 0.1

(p.228) (p.229) The correlation of the distribution of the ST with that of the general population across states is much lower than of the SC; therefore one might expect the effect of trade liberalization and openness to be different for the former. Indeed, this is very much in keeping with what we find. In general, the statistical significance of the openness variables is much lower for the ST. In particular, reductions in nontariff barriers seem to have no significant effect on poverty. Reductions in tariffs in urban areas are associated with a large reduction in poverty (4.1 percent) and are also statistically significant at a high threshold, but recall that the ST population in urban areas is tiny. In rural areas, where the ST population is concentrated, openness measures do not have a statistically significant effect.

Table 5.12 Poverty and Trade Openness (Non-Scheduled Individuals)

Combined

Urban

Rural

NRP

NTB

FPC

NRP

NTB

FPC

NRP

NTB

FPC

VARIABLES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Protection

.0066*

.0262*

.326**

.0223**

0.02

1.1041**

0.01

.0375**

0.29

[.004]

[.013]

[.134]

[.009]

[.023]

[.454]

[.005]

[.018]

[.175]

Dev. Exp. Pc

0.06

0.12

0.06

0.72

0.68

0.64

0.27

0.33

0.25

[.262]

[.257]

[.251]

[.496]

[.543]

[.495]

[.363]

[.339]

[.353]

Constant

2.03

–0.4

2.11

–3.85

–2.83

–2.59

0.31

–3.37

0.42

[1.747]

[2.191]

[1.677]

[3.514]

[4.095]

[3.365]

[2.409]

[2.871]

[2.342]

Observations

45

45

45

45

45

45

45

45

45

R 2

0.96

0.96

0.97

0.92

0.9

0.92

0.94

0.95

0.94

Standard errors in square brackets

*** p 〈 0.01, ** p 〈 0.05, * p 〈 0.1

Finally, for completeness, we also carry out the analysis for the NS as a group. The results indicate that the relationship between poverty reduction and trade liberalization for those in the NS group closely mirrors those for the population as a whole. A percentage reduction in tariffs is associated with a .65 percent fall in poverty. Similarly, a 1 percent reduction in nontariff barriers leads to a 2.5 percent fall in poverty. In rural areas, the effect of nontrade barriers rises to 3.8 percent while it is insignificant in urban areas. On the other hand, a fall in tariffs is associated with a 2.2 percent fall in poverty in urban areas and has little effect on rural poverty.

In summary, our results find strong support for the positive relationship between trade liberalization and overall poverty reduction in India between 1983 and 2004. These results vary substantially across urban and rural areas. Most importantly, from the viewpoint of our focus on social groups, we find that poverty reduction for the NS and SC tracks that for the population as a whole reasonably closely, but the effect is weaker for the ST. In no case, rural or urban and SC, ST, or NS, do we find the effect of increased openness on poverty to be positive and statistically (p.230) (p.231) (p.232) (p.233) significant. In this sense our results strongly support those of Hasan et al. (2006–7) and Cain et al. (2010) and refute those of Topalova (2007, 2010) for the population as a whole. On balance, the rising tide seems to have lifted all boats, even if some have been lifted more than others. Most importantly, it has lowered, let alone sunk, none.

Sensitivity Analysis

Two characteristics of our data call for sensitivity checks. First, the year 1987–88 was subject to a drought that had a differential impact across states. In all probability, some of the states with large populations of the poor, SC and ST, were subject to greater adverse impact than others. Therefore, the observations from year 1987–88 may not follow the same pattern as those coming from other years. Second, in many states, our poverty estimates for the ST are based on relatively small number of ST households sampled by the survey. These estimates are likely to be imprecise and may bias our estimates of the relationships between poverty and per capita income and poverty and openness. In this section, we check the implications of correcting for these sources of biases.

In order to check whether the drought in 1987–88 may have contaminated the relationship between poverty and per capita income, we rerun equation (2) after excluding observations from the 43rd round conducted in 1987–88. We report the resulting coefficients and associated standard errors in Table 5.13. As can be seen, other than minor differences in the magnitudes, the coefficients and their statistical significance are virtually the same as in Table 5.8. We conclude that the inclusion of observations from the drought year of 1987–88 did not bias the direction of our coefficients and had marginal effects on the magnitudes.

Table 5.13 Per Capita Income and Poverty (excluding 1987–88)

VARIABLES

SC

ST

NS

All

Ln pc

–5.35

–15.8489**

–10.9272***

–9.2894**

[4.425]

[7.329]

[3.326]

[3.464]

Lagging state

–9.7863**

7.05

–1.15

–1.4

[4.627]

[7.663]

[3.478]

[3.622]

Landlocked

10.6108***

–17.7410***

0.73

1.36

[3.865]

[6.401]

[2.905]

[3.025]

SC initial

0.5436**

0.7869*

0.21

0.4764**

[0.237]

[0.392]

[0.178]

[0.185]

ST initial

0.1

0.9275***

0.12

0.3599**

[0.185]

[0.306]

[0.139]

[0.144]

Poverty 73–74

1.2993***

0.27

0.6955***

0.7808***

[0.177]

[0.293]

[0.133]

[0.138]

Year 1993

–8.2817**

1.14

–5.6328**

–5.9540**

[3.144]

[5.207]

[2.363]

[2.461]

Year 2004

–17.1131***

–5.01

–9.4050***

–10.9078***

[4.214]

[6.979]

[3.167]

[3.299]

Constant

15.82

138.0587**

77.3844***

58.9949**

[34.370]

[56.927]

[25.835]

[26.904]

Observations

54

54

54

54

R2

0.75

0.64

0.8

0.81

Standard errors in square brackets

*** p 〈 0.01, ** p 〈 0.05, * p 〈 0.1

(p.234) The implications of dropping the observations that are based on small number of ST households are more mixed. A key problem in approaching this question is that we have no objective basis for defining a “threshold” number of ST households below which the poverty estimates are to be considered imprecise. As Appendix Table 5.A1 shows, the number of ST households picked up by the survey varies widely across states and survey years. As a result, whether we draw the line at 100, 50, or 30 households causes a considerable difference in the number of states that get included in the regressions.

We alleviate this problem by taking two alternative approaches to identifying the states to be excluded from the regression sample. First, we test whether the poverty estimate for a state in a given year can be identified as being statistically significantly different from that in any of the other years for which a sample survey was conducted. The idea here is to check whether the estimate is tight enough to be distinguished from another estimate in a statistical sense. The criterion is, of course, not foolproof, in that if the two poverty estimates being compared are indeed equal, the test will accept the null hypothesis of equality even when each of the estimates is based on a very large number of households. Conversely, if the estimates are vastly different, there is a possibility that the null will be rejected even when the two estimates are based on a small number of observations. Our second approach is to require the poverty estimate to be based on an arbitrarily chosen minimum number of ST households to be included in the sample used for the regressions.

When we consider the combined (rural-plus-urban) sample, the first criterion excludes only one observation: Delhi in 2004–5. Without reporting the estimated equations, we note that this exclusion does not change our results with respect to either the impact of per capita income or openness on poverty. Indeed, since Delhi was absent from the regression sample for openness and poverty (Tables 5.95.12) in (p.235) (p.236) the first place due to unavailability of protection measures for it, this regression remains unchanged.

The criterion is more discriminatory when we consider rural and urban samples separately. It reduces the number of observations to thirty-three for the urban sample and forty-two for the rural sample. Without reporting the regression results, we note that they remain qualitatively unchanged from Table 5.11. The two measures of protection (NRP and FPC) remain positively and statistically significantly associated with urban poverty, while protection shows no statistically significant effect on rural poverty for the ST.

Our second criterion leads to more serious deviations from our original results. For the combined (rural plus urban) sample, we set a threshold of a minimum sample of seventy or more ST households for inclusion in the regression. This leads us to drop Delhi, Haryana, J&K, Kerala, and Punjab from all rounds and Tamil Nadu from the 61st round. The change once again leaves the results in Table 5.7 largely unaffected, but those in Table 5.8 are altered: the coefficient of per capita income becomes positive and statistically insignificant.

Turning to the impact of openness on poverty, when we drop the observations based on a minimum number of ST households, surprisingly, our results are considerably strengthened. Keeping the trade-off between the number of households necessary to get a (p.237) reasonably precise estimate of poverty and the number of observations necessary to get a meaningful regression fit, we proceed as follows:

  1. (1) For the regressions involving combined samples of rural and urban households, we exclude all observations involving less than seventy households. This leads to the exclusion of Delhi, Haryana, J&K, Punjab, and Kerala from all rounds and Tamil Nadu from the 61st round.

  2. (2) For the regressions involving rural households, we exclude Delhi, Haryana, J&K, Punjab, Kerala, and Tamil Nadu from all rounds. This ensures that the poverty estimates for the included states are based on seventy or more ST households.

  3. (3) For the regressions involving urban households, the exclusions are much larger, since the surveys pick very small number of ST households in many states. Here we drop Delhi, Haryana, J&K, Punjab, Himachal Pradesh, Tamil Nadu, Rajasthan, and UP from all rounds. This list still includes two states with fewer than seventy observations, but cutting the list any further would have left us with too few observations.

Taking into account the fact that we have protection measures for only fifteen states and three survey rounds, the above restrictions reduce the sample size to thirty-five for the combined (rural plus urban) sample, thirty-three for the rural sample, and twenty-nine for the urban sample. Our results are reported in Table 5.14. Comparing these results to those in Table 5.11, we see that dropping the observations associated with poverty estimates based on small number of ST households strengthens the results. At least one measure of protection now leads to a positive and statistically significant effect on poverty for rural, urban, and combined poverty measures. In the case of (p.238) urban poverty, each of the three measures of protection has a statistically significant effect.

Table 5.14 Openness and Poverty for the ST (Robustness Tests)

Combined

Urban

Rural

NRP

NTB

FPC

NRP

NTB

FPC

NRP

NTB

FPC

VARIABLES

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Protection

0.01

.1133***

0.79

.0287***

.0827**

1.4083***

0.01

.140***

0.8946

[.013]

[.037]

[.479]

[.008]

[.031]

[.365]

[.014]

[.035]

[.527]

Dev. Exp. Pc

.8039*

.9021**

.7705*

0.38

0.15

0.3

1.1937**

1.1987***

.9978**

[.401]

[.329]

[.379]

[.566]

[.624]

[.537]

[.480]

[.329]

[.452]

Constant

–2.58

–13.70***

–2.81

0.27

–5.34

4.64

–5.08

–18.423***

–2.063

[2.747]

[4.390]

[2.559]

[4.173]

[5.004]

[4.029]

[2.995]

[4.016]

[3.950]

Observations

35

35

35

29

29

29

33

33

33

R 2

0.83

0.88

0.84

0.96

0.96

0.97

0.85

0.92

0.871

Standard errors in square brackets

*** p 〈 0.01, ** p 〈 0.05, * p 〈 0.1

Concluding Remarks

We begin the concluding remarks with an emphatic statement on the bottom line of what we have shown: there is absolutely no statistically significant evidence that rising incomes and increased openness have negatively impacted any of the three broad social groups. This does not preclude the possibility that specific individuals within these groups, including those belonging to the Non-Scheduled group, may fail to enjoy any benefits. But we find no evidence of any harmful impacts at the average level for each social group.

Beyond this bottom line, our chapter has offered the most up-to-date and comprehensive analysis of poverty among the socially disadvantaged groups in comparison to those without social handicaps. We offer compelling evidence of steadily declining poverty among all groups in rural as well as urban areas at the national level. We also provide the trends in poverty ratios in the top ten states by the SC and ST populations. The top ten SC states account for 85 percent of the total countrywide SC population, and the top ten ST states for more than 90 percent of the countrywide ST population. In the case of the SC, every one of the top ten states saw the poverty ratio decline between 1993–94 and 2004–5. This is not true of the ST states, however, where reversals were observed in some states.

We demonstrate that declining poverty rates are uniformly associated with rising per capita incomes. In the case of the ST, the Non-Scheduled, and all groups taken together, we show that rising per capita incomes have a statistically significant and negative effect on poverty at the 95 percent or higher level. Our sensitivity tests (p.239) (p.240) dilute slightly the result for ST poverty, but do not alter our basic message. We also find that rising per capita incomes are accompanied by declining poverty rates within the SC, but the estimated coefficient is statistically insignificant at the 90 percent or higher level.

Finally, we also conduct an econometric test of the effect of openness on poverty for various social groups. In the case of the SC and NS, we find that one or more measures of openness have a statistically significant and negative effect on poverty in rural and urban areas as well as the two areas combined. In the case of the ST, the coefficient is statistically significant in urban areas only when we use the entire sample. But once we exclude the observations associate with poverty estimates based on small numbers of ST households, the results for the ST are further strengthened. After this correction, we find that at least one measure of protection is associated with a statistically significant increase in poverty in rural, urban, and combined regions and, moreover, that all three measures have a positive and significant effect on poverty in the urban regions.

(p.241) Appendix

Appendix Table 5.A1 Number of ST Households by State and by Surveys

State

61st Round

50th Round

43rd Round

38th Round

Rural

Urban

Rural

Urban

Rural

Urban

Rural

Urban

Andhra Pradesh

446

80

373

85

390

73

350

60

Assam

674

110

496

46

497

75

641

17

Bihar

35

7

557

109

979

93

806

109

Chhattisgarh

651

135

Dadra and Nagar Havel

137

13

205

34

274

239

Delhi

0

16

2

18

1

23

4

19

Goa

2

0

3

5

33

25

4

Gujarat

443

129

483

112

544

135

532

74

Haryana

6

5

18

6

22

11

7

8

Himachal Pradesh

149

10

107

10

93

14

85

14

Jammu and Kashmir

13

7

15

13

31

18

5

7

Jharkhand

665

124

Karnataka

207

63

212

96

168

73

237

56

Kerala

54

10

37

17

55

16

56

14

Madhya Pradesh

863

157

1,561

202

1,896

207

1,939

175

Maharashtra

577

208

563

202

719

227

820

201

Orissa

908

121

816

111

765

144

736

100

Punjab

8

10

32

15

25

25

29

15

Rajasthan

530

64

483

46

514

53

494

50

Tamil Nadu

15

20

82

43

42

31

45

54

Uttar Pradesh

40

26

83

27

81

105

156

35

Uttarakhand

74

14

West Bengal

366

48

374

59

394

62

417

63

Total

6,863

1,377

6,606

1,291

7,647

1,435

7,694

1,113

(p.242) (p.243)

Notes

(p.246) References

Bibliography references:

Cain, Jewel, R. Hasan, and D. Mitra (2010). “Trade Liberalization and Poverty Reduction: New Evidence from Indian States.” Working Paper no. 2010-3, Columbia Program on Indian Economic Policies, School of International and Public Affairs, Columbia University.

Das, D. (2008). “Trade Barriers Measurement: A Methodological Overview.” Mimeo.

Dubey, Amaresh, and S. Gangopadhyay (1998). Counting the Poor: Where Are the Poor in India? Sarvekshana Analytical Report no. 1. New Delhi: Ministry of Statistics and Program Implementation, National Sample Survey Organization.

Hasan, R., D. Mitra, and B. Ural (2006–7). “Trade Liberalization, Labor Market Institutions, and Poverty Reduction: Evidence from Indian States.” India Policy Forum 3, 70–135.

Hnatkovska, Viktoria, Amartya Lahiri, and Sourabh B. Paul (2010). “Castes and Labor Mobility.” Department of Economics, University of British Columbia, October.

Kijima, Yoko (2006). “Caste and Tribe Inequality: Evidence from India, 1983–1999.” Economic Development and Cultural Change 54(2), 369–404.

Meenakshi, J. V., R. Ray, and S. Gupta (2000). “Estimates of Poverty for SC, ST and Female-Headed Households.” Economic and Political Weekly, 35(31), July 29, 2748–54.

Panagariya, Arvind (2008). India: The Emerging Giant. New York: Oxford University Press.

Pandey, Mihir (1999). NCAER Report on Trade Protection in India. National Council of Applied Economic Research, New Delhi, India.

Shukla, Rajesh, Sunil Jain, and Preeti Kakkar (2010). Caste in a Different Mould. New Delhi: Business Standard.

Sundaram, K., and S. Tendulkar (2003). “Poverty among Social and Economic Groups in India in 1990s.” Economic and Political Weekly 38(50), December 13, 5263–76.

Topalova, Petia (2007). “Trade Liberalization, Poverty and Inequality: Evidence from Indian Districts.” In Ann Harrison (ed.), Globalization and Poverty. Chicago: University of Chicago Press, 291–336.

Topalova, Petia (2010). “Factor Immobility and Regional Impacts of Trade Liberalization: Evidence on Poverty from India.” American Economic Journal: Applied Economics 2(4), 1–41.

Notes:

(*) The authors are, respectively, at the London School of Economics and Columbia University and can be reached at m.mukim@lse.ac.uk and ap2231@Columbia.Edu. They thank Rajeev Dehejia, Devashish Mitra, and Rana Hasan for extremely helpful interactions in the course of writing this chapter and K. Sundaram for his excellent comments at the Columbia-NCAER conference held in New Delhi on March 31–April 1, 2011. Work on this chapter has been supported by Columbia University’s Program on Indian Economic Policies, funded by a generous grant from the John Templeton Foundation. The opinions expressed in the chapter are those of the authors and do not necessarily reflect the views of the John Templeton Foundation.

(1) . The SC and ST classifications have their origin in the Indian Constitution, which lists in separate schedules castes and tribes that are officially recognized as having suffered discrimination for centuries and therefore requiring special affirmative action. A program of reservation in public sector jobs and public schools, colleges, and universities for the SC and ST in India has existed since the early 1950s. Reservation in government jobs was extended to the OBC in 1990 and in central government educational institutions in 2006.

(2) . The document containing the statement by the National Campaign on Dalit Human Rights is posted at http://www.publications.parliament.uk/pa/cm201011/cmselect/cmintdev/writev/616/m02.htm (accessed June 3, 2011). The statement by Bidwai can be found in the article “Equity, Not Growth, Is the Key” posted at http://www.mydigitalfc.com/op-ed/equity-not-growth-key-359 (accessed June 3, 2011).

(3) . A recent study by Shukla, Jain, and Kakkar (2010) also focuses on the prevailing inequality among various social groups, offering rich set of indicators drawn from the National Survey of Household Income and Expenditure conducted by the National Council on Applied Economic Research. Additional references to earlier studies on inequalities between Scheduled and Non-Scheduled groups can be found in the reference lists in Kijima (2006) and Hnatkovska, Lahiri, and Paul (2010).

(4) . An earlier paper by Meenakshi, Ray, and Gupta (2000) provided statewise poverty estimates for the SC and ST for a single year, 1993–94, but did not deal with the issue of the evolution of poverty among these groups over time and how it may have been impacted by growth and reforms. These authors also refer to Dubey and Gangopadhyay (1998) as having provided the estimates of poverty by social groups in 1993–94 and perhaps earlier years.

(5) . Thick surveys are typically in the field from July 1 of the beginning calendar year to June 30 of the ending calendar year. For example, the 2004–5 survey was in the field from July 1, 2004, to June 30, 2005. The 38th round followed a different schedule, staying in the field from January 1 to December 31, 1983. (p.245) India has had a long tradition of conducting sample surveys; the first expenditure survey took place as early as the period from October 1, 1950, to March 31, 1951.

(6) . The poverty rate for each of these entities is calculated by recourse to one or another special assumption. For the six northeastern states and Sikkim, the poverty ratio is assumed to be the same as for Assam. In the case of Goa, its expenditures are combined with the poverty line of Maharashtra to calculate the poverty ratio. Similar assumptions are made for the UTs.

(7) . Targeted assistance is often provided based on the location of individuals. Rural employment guarantee schemes or subsidized food prices available to the rural poor, for instance, have this attribute. Such policies inadvertently impede migration, which may be detrimental to development in the long run.

(8) . For lagging and leading regions, we use the definition by the World Bank, which classifies the states into these regions according to per capita income. With respect to the 1973–74 poverty levels, ideally, we would like to have the proportion of the poor within SC population for SC regression, within ST population for ST regression, and so forth. Unfortunately, we do not have unit-level data for the quinquennial survey conducted in 1973–74 to carry out these calculations. The aggregate poverty rates are available from the Planning Commission website (http://planningcommission.gov.in/data/datatable/), however.

(9) . These coefficients remain significant if we replace the 1973–74 poverty levels with those from 1977–78, but the level of significance declines: upon replacement, the coefficients of SC and all groups pass the 90 percent threshold but not the 95 percent or higher threshold, while that of NS continues to pass the 99 percent threshold.

(10) . Cain et al. take the industry-level tariff rates and NTB coverage rates between 1988 and 1998 from Pandey (1999) and those for the subsequent years, until 2003, from Das (2008). The latter constructs the rates using the same methodology as the former. Because these sources do not provide the protection rates for every single year between 1988 and 2003, Cain et al. use simple linear interpolation to obtain the relevant rates. The rates for 1986, necessary to include the data from the 1987–88 survey, are estimated by assuming that tariff and NTB coverage rates grew at the same annual rate from 1986 to 1988 as they did from 1988 to 1989. The NTB coverage rates are of course bounded at 100 percent.

(11) . Cain et al. choose employment weights from the year 1993–94 because it is one of the middle years in the data and is therefore a good candidate to serve as the base (reference) year in the construction of state-level openness index. As with any good index, the weights are not held fixed over time.

(12) . Hasan et al. (2006–7) experimented with contemporaneous protection but it did not affect the results in any substantive manner. As such, we work exclusively with lagged protection measures here.