Document(s) 2 of 16

This chapter ¹ attempts to assess the impact of the economic reforms (including liberalization) on the incidence of poverty in India. It does so by comparing the changes over the period 1993–1994 to 1999–2000 with the course of poverty decline in the previous quinquennial 1987–1988 to 1993–1994. These are the two periods covered by the six yearly ('thick') rounds of the National Sample Survey (NSS) household expenditure surveys.

A number of researchers have already worked on these particular data sets. Major contributions have been published by Sundaram and Tendulkar (2000, 2001 and 2003) and by Deaton and Dreze (2002). Our work builds on and extends these important pieces of research.

In the first section of the chapter we outline the methodology of decomposition changes in the incidence of poverty used elsewhere by Mazumdar and Son (2002). It seeks to quantify the components of the change between any two periods between those due to growth in mean income (or expenditure), those due to distribution of income, and those due to a shift of population between sectors of the labor market (e.g., the rural and the urban sectors). The first section also undertakes a reassessment of the data on measured poverty for the NSS rounds in question. As is well known from the work referred to above, there are two important issues in the empirical use of the NSS data. First, we have the problem of the correct assessment of the poverty line for different sectors, dates and regions of India. Second, we have the problem of comparability of the measured household expenditure per capita over time due to some changes in the recall period in the successive NSS surveys. In the first section we extend the Deaton and Dreze method of assessing the poverty lines, and the Sundaram–Tendulkar approach to deriving consistent estimates of the average per capita expenditure (APCE) of households at different dates. New estimates of the incidence of poverty are presented.

The second section discusses the main results for all-India. An important extension to previous work is the explicit distinction made in the analysis between metro and non-metro towns.

The third section contains a detailed analysis of the state-level data, and some progress is made in the unraveling of critical inter-state differences in the changes in the incidence of poverty over the periods studied.

The final section presents the major conclusions of the work.

The decomposition of poverty

The conventional method of understanding the dynamism of changes in poverty is done through various inequality measures including that of Lorenz curve and various Entropy measures. The decomposition exercise undertaken here does not require us to specify an inequality measure. It uses an idea of shift in that part of the Lorenz curve, which affects the poor.

A brief exposition of decomposition methodology

To get better understanding about dynamics of changes in poverty, the change in the incidence of poverty can be broken down into three elements: (i) any shift in population between the different segments with different degrees of poverty; (ii) the growth in income in each of the segments; and (iii) the change in the distribution of income, particularly at the lower end where the poor households are located. The methodology of such decomposition is set out in Appendix 1. To summarize the result, the percentage change in poverty for the whole economy can be expressed as:

where f_i and P_i are the population share and poverty index of the ith group respectively;

the subscript m denotes the change in poverty due to mean income growth and the subscript I gives the measure of poverty change due to change in inequality.

The first term in equation (1) measures the effect of growth within each group on overall change in the poverty incidence, when the distribution within each group remains the same over time. This first term can be further decomposed into two terms:

The first term on the right-hand side of above equation measures the effect of growth on percentage change in poverty under the counter-factual that all groups enjoyed the same uniform growth rates and the second term in the right-hand side measures the effect of differential growth rates within groups.

So substituting (2) into (1), we get a decomposition that expresses the percentage change in the poverty incidence as the sum of four components: (1) overall growth effect when inequality in the distribution does not change; (2) effect of differential growth rates in different groups; (3) effect of change in inequality within different groups; and (4) effect of changes in population shares between groups. This is an exact decomposition and, therefore, there will not be any residual term.

The database

The database used in this analysis is the 'consumption expenditure survey' of various quinquennial rounds of National Sample Survey Organisation (NSSO).² The purpose of this study is a comparison of the incidence in poverty between the pre- and post-liberalization periods in India. We have done our analysis on the basis of the three quinquennial rounds, i.e., 43rd (1987–1988), 50th (1993–1994) and 55th (1999–2000). We intend to capture the impact of liberalization by comparing the change in poverty for the first period (1987–1988 to 1993–1994) with that in the second period (1993–1994 to 1999–2000). The decomposition exercise for the Head Count Ratio (HCR), the poverty gap ratio (PGR) and the squared poverty gap ratio has been carried out for 16 major states and for all India (16 states combined).

Adjustment made in the database

We have used average monthly per capita consumption expenditure (APCE) as the proxy for per capital income. However, certain adjustments were made to APCE for the year 1993–1994. Expenditures of all consumer items of 1987–1988 and 1993–1994 are based on a 30-day recall period, known as uniform recall period (URP). For 1999–2000, all but five items are based on the 30-day recall period. The expenditures on the five remaining items are based on 365-day recall periods. These items are clothing, footwear, education, institutional medical expenses and consumer durables. So for 1999–2000, the reference period is known as the mixed reference period (MRP). In the year 1993–1994, for these five consumer items, expenditure data were collected for both 30-day and 365-day reference period. To make 1993–1994 data comparable with 1999–2000 data we replaced 30-day expenditures of these five items with 365-day expenditure.³ In this fashion, we could change 1993–1994 URP (uniform reference period) expenditures into MRP (mixed reference period) expenditure. A comparison of APCE on these five items by the 30-day and the 365-day reference periods for the year 1993–1994 showed that in both rural and urban areas the change of the reference period from 30-day to 365-day made substantial difference largely in clothing.

We could not convert 1987–1988 consumer expenditure data into MRP in a similar fashion because the expenditure data on the above-mentioned five items were not collected for both the 30-day and the 365-day reference periods. Hence the decomposition analysis for changes in poverty for the period 1987–1988 to 1993–1994 (pre-liberalization period) will be based on URP and for the period 1993–1994 to 1999–2000 will be based on MRP. We will surely lose continuity of poverty estimates in this fashion but avoiding this important issue would otherwise lead to an upward bias in the reduction of poverty in the post-liberalization period relative to the pre-liberalization one.

It might be objected that flow expenditure on low-frequency articles like durables might be reported differently by poor people than by rich consumers for the two alternative recall periods. In fact experiments performed in the 'thin' samples of the 51st to the 54th rounds showed that on the 365-recall period, lower-income households reported higher annual rates compared to the 30 day recall method, but richer households had exactly the opposite bias. At the same time there is an expectation that there has been a large increase on durables affecting all classes. A comparison of the change in APCE based on the 30-day recall as for the first period might not be strictly comparable to the change in the second period based on the 365-day recall. One might be missing less at the mean in the first change than in the second change.⁴ An examination of the detailed data on consumption by items and income groups, however, showed that the major difference for poorer groups in the reported expenditure by the two recall periods was in clothing, not in all durables. In the lowest eight income groups, ranging up to the 35–40 fractile, in the 1993–1994 (50th round) survey, the highest difference was Rs.2.5 for durables compared with Rs.17.25 for clothing (the full data are given in Mazumdar and Sarkar 2004). Thus the income-related bias in reported flow of expenditure on durables might not quantitatively of great importance in the two periods of comparison with different recall periods.

Choice of poverty line

In choosing the poverty line we deliberately did not choose the official poverty line as given by the Planning Commission of India. Historically, the rural–urban price differential as incorporated in the official poverty lines at all-India level was around 15 percent level. But the 1993 Expert Group Report recommended separate rates for each state (based on studies of interstate price differentials) and did not explicitly consider the urban to rural differentials. As a result, in 1999–2000, the urban to rural differential implicit in the official lines was around 39 percent and it is astonishingly large for some states (Deaton 2003). The effect of the adoption of the Expert Group lines was to raise measured poverty in urban relative to rural areas. The poverty-line figures, by state and sector, are calculated by using the Tornqvist price index presented by Deaton (2003).⁵

Following Deaton's procedure, the starting point for calculation of poverty indexes is the official rural all-India poverty line for the 43rd round, 1987–1988. The figure is Rs.115.7 per capita per month. First, rural poverty lines for states are obtained by multiplying this base poverty line by rural price indexes for each state relative to all India. Urban poverty lines, for each state as well as for all-India, are calculated from the rural poverty lines by scaling up by the respective urban relative to rural price indexes.

Similarly, for the 50th round (1993–1994), the all-India rural poverty line of 115.7 of the 43rd round is scaled up by the index for all-India rural for the 50th round relative to the 43rd round. The figure thus calculated is 196.5. Rural poverty lines for states are obtained by multiplying this base poverty line by rural price indexes for each state relative to all-India. Urban poverty lines, for each state as well as all-India, are calculated from rural poverty lines by scaling up by the respective urban relative to rural price indexes (see Mazumdar and Sarkar 2004 for the table of poverty lines by states and rural–urban location).

Results for all-India

We first discuss the pattern of poverty decline in the two periods for the whole of India–based on the new figures for the 16 states considered. The results for the decomposition analysis are presented in Table 2.1.

According to our estimates the head-count ratio (HCR) was reduced at a perceptibly higher rate in the more recent period–the decline was about 20 percent higher. This apparent acceleration is, however, largely due to the smaller base of the HCR at the beginning of the second period. The absolute decline in HCR was 6.3 percentage points in 1987–1993 and 5.3 in the 1993–2000 periods. Thus our figures support the conclusion of Deaton and Dreze that 'poverty decline has been fairly evenly spread between the two sub-periods (before and after 1993–1994) in contrast with the pattern of acceleration in the second sub-period associated with the official estimates' (p. 3734).

Growth of mean consumption accelerated in the second period, and played a larger role in the poverty reduction in this period. It can be seen that the inequality effect overall (i.e., taking the rural and urban areas together) continued to play a contributory role to poverty reduction, but the share of this factor in the reduction was much reduced.

Important changes, however, emerged in the relative importance of the rural and the urban areas in the process of poverty reduction. The share of the urban areas in the overall poverty decline increased in the later period (from 12 percent of the total percentage decline to 21 percent). This bigger role of the urban sector in poverty decline was, however, not due to accelerated population shift to the urban sector. In fact, the 'population shift' effect, while playing a minor role in both periods, actually decreased significantly in the 1993–2000 period.

The crucial element was the higher growth rate in the urban sector. Both the sectors increased the rate of mean growth, but it can be seen from the third column of Table 2.1 that the differential effect of growth rates reduced the incidence of poverty significantly in the urban areas, but increased it in the rural areas. If the latter had grown at the same rate as the rest of the economy, poverty reduction in the 1993–2000 periods would have been 30 percent higher.

The impact of the differential growth rate was balanced to some extent by the impact of changes in inequality in the two sectors. While inequality (in the relevant range of the Lorenz curve) decreased in the rural sector, it deteriorated somewhat in the urban sector, thus canceling out some of the poverty-reduction effect of the differential growth in the sector. But the inequality effects were not as strong as the differential growth effect. This explains the larger role of the urban areas in poverty reduction in this period.

Elasticity of poverty decline and the poverty gap ratio

We have seen that both the growth rate of mean consumption and the rate of decline in the headcount ratio of poverty accelerated in the second period of our study. The elasticity of the change in HCR with respect to the growth in consumption is of interest. Table 2.2 shows the numbers for the two periods. The values of the elasticity in the rural and urban areas are very close together. In both sectors there has been a significant fall in the elasticity in the post-reform period. The results of the decomposition analysis given in Table 2.1 suggest that the reasons for this decline are different in the two sectors. In the rural economy the inequality effect increased its negative value, suggesting that ceteris paribus the effect on HCR of growth would be strengthened. But there was a significant fall in the shift of the labor force out of this sector, which weakened the impact on poverty. By contrast the urban areas–where the change in HCR benefited from the fall in the intake of labor–suffered from an adverse distributional effect.

The HCR of course only measures the number of people below the poverty line, and does not take account of the economic distance of the poor from the poverty line. This is addressed in the group of measures called the 'poverty gap ratio' (PGR) and its variants. Table 2.3 shows the values of the elasticity of

PGR with respect to mean consumption growth in the two periods. It is seen that the elasticity of the PGR with respect to growth is much higher than that of the HCR in both periods and both sectors. This partly reflects the fact that the absolute value of the initial PGR is much lower than that of the HCR. But it also draws attention to the substantive point that the growth process has affected people further below the poverty line strongly–not just those slightly below the threshold.

A second interesting point to note is that as with the HCR the elasticity of PGR fell in the second period in both sectors. However, the decline in elasticity in case of PGR seems to have been much stronger in the rural areas suggesting that the poorest of poor were worse hit.

Table 2.4 gives the results for the decomposition analysis for the poverty gap measure. The different components of the poverty reduction appear to behave in much the same way for this measure as for the HCR. As already pointed out the percentage decrease in PGR is larger than in the HCR, but qualitatively the role of all three components of poverty decline is the same in the two cases. In the post-reform period the role of population shift is much reduced. The urban areas gain in the strength of the growth effect, but inequality increases, offsetting the effect of growth to some extent. The relative slowdown in growth in the rural sector is partly countered by a favorable inequality effect.

There are differences in the relative magnitudes of the various effects. One interesting difference is that in the recent period the inequality effect on the PGR seems to be stronger than on the HCR in the rural areas, but the other way round in the urban sector. In the rural sector the inequality effect is 38 percent of the growth effect in the HCR decomposition, but 52 percent in the PGR analysis. The offsetting effect of increased inequality is, however, weaker for the PGR in the urban areas (15 percent of the growth effect as against 25 percent for the HCR decomposition).

We are inclined to agree with the Deaton–Dreze conclusion that very little additional insight is to be gained from the detailed analysis of the poverty- gap ratio or its further refinements over and above what we learn from the simple analysis of the HCR (Table 2.5). In view of this we will make no further reference to the measures other than HCR in the subsequent discussions.

Metro and other urban areas

An interesting question pertains to the relative importance of metropolitan (population > one million) and other urban areas in poverty reduction.

The decomposition of HCR was done for metro and other areas separately. Out of the 16 major states considered for the decomposition analysis, only seven states had a metro city in the year 1987–1988, ten states in the year 1993–1994 and 11 states in the year 1999–2000. So to maintain uniformity, we analyzed seven states separately that had a metro city throughout our period of analysis. However, separate analysis was undertaken for the three states that had a metro area only since 1993–1994.

The incidence of poverty is as expected higher in the non-metro areas. We studied changes in the incidence of poverty over the two sub-periods between 1987–1988 and 1993–1994, and 1993–1994 and 1999–2000 and these have been worked out on the basis of both the URP and the MRP criteria (Mazumdar and Sarkar 2004, Table II.3).

For the seven states the decline in poverty incidence in absolute terms does not differ much between metro and non-metro areas in the 1987–1993 period. This implies that in proportionate terms the decline is much more in the metro areas. But the trend seems to have been reversed in the later sub-period.

The absolute decline is much smaller in the metros as the HCR seemed to be nearing the floor level, though the difference in percentage terms is not all that much.

Our decomposition analysis was applied to the data for the metro and the non-metro areas in the same way that we had done for the rural and the urban areas as a whole. The results are set out in Table 2.6.

It is seen that a major change in the more recent period was registered by the non-metro urban sector. Differential growth rate favored poverty reduction in a more pronounced way in the non-metros. The non-metro sector also suffered relatively much more from an increase in inequality which seemed to have affected the urban areas as a whole. Combined with a more positive contribution from population shift to poverty reduction, almost the entire decrease in urban HCR in the 1993–2000 period was accounted for by the non-metro sector.

The urban sector by size classes of towns

For some purposes it might be better to classify the urban sector by more size classes than just two. We distinguished three sub-groups: towns with a population of less than 50,000 (small); those larger than this but with less than ten lakhs or one million (medium and large); and those more than one million (metro).

For the country as whole, there is a remarkable difference between the two periods. For the 1987–1993 period the rate of growth of APCE was directly related to the size of towns, the largest towns having the highest growth rate. Accordingly the rate of decline in HCR was also varied directly with the size groups–and in fact this positive relationship was much stronger. In the post-reform years 1993–2000 the relationship has been reversed remarkably. The reversal again is much more prominently revealed in the variations in the rates of decline of the HCR. The small towns had a rate of decline 50 percent higher than the largest group.

This interesting result poses the question: what aspect of the post-reform growth process has been responsible for this reversal of the fortunes of the small towns relative to the larger ones? It is possible to hypothesize that the result might be the consequence of either a strong 'trickle down' effect powered by a decentralization of non-agricultural activities in the urban sector. Alternatively the smaller towns might have enjoyed a stronger growth rate (and poverty reduction) in the post-reform years because of the impact of the growth and commercialization of the agricultural economy. It is probable that both influences have been present in the process observed.

It should be noted that some of the individual states show trends different from the one just mentioned for all-India. There are, in particular, five states in which the rate of growth of APCE and HCR are directly related to the size class of towns–the opposite of the result for India as a whole. These are: Gujarat; Karnataka; Maharashtra; Rajasthan; and Tamil Nadu. As it happens, these are the states which have been the largest recipients of Foreign Direct Investments. Since it is well known that FDI goes almost exclusively to metro cities, the hypothesis suggested strongly is that FDI has given an uplift to the growth rate of metro areas in those states where it has played a significant role–and that this impact has raised the growth rate of mean consumption sufficiently induce a higher rate of poverty decline than would be expected by looking at the average for all-India.

The relevant data on FDI per capita in metros by states are plotted along with the rate of growth of APCE in the individual states for the 1993–2000 period in the scatter diagram of Figure 2.1. The relationship is found to be a very strong one.

Figure 2.1 Relationship between FDI and growth rates of APCE in metro areas.

Note

See full names of different states like WB, MP, AS, etc. in first column of Table 2.8.

This result suggests that there are two different aspects to the impact of the post-reform developments including globalization on growth and poverty decline in the urban metro sector. On the one hand, there has been a distinct trend towards decentralization of economic activities to smaller towns and cities. This has led to the inverse relationship between growth and the size of towns observed in many states and in India as a whole. There are, however, a group of states in which the role of FDI is strong, and the impact is seen in a high growth rate in metro areas, so that a direct relationship between growth (and poverty reduction) and town size is observed. The only exception to this two-way classification of states is Rajasthan. The inflow of FDI per capita in metros in this state is low, yet it shares the characteristics of the high FDI states in having a relatively high growth rate in the larger towns (see Figure 2.1). It is, however, well known that even if the inflow of FDI is small Rajasthan has participated in the globalization process strongly through the promotion of international tourism in particular.

The implication of the argument of the last paragraph is that the 'trickle down' effect on smaller towns has been more important in the states with a lower level of international connection. The reform process has encouraged decentralization in these states. But what about the impulse to growth in small towns coming from the rural sector? Figure 2.2 plots the growth rate of APCE in small towns (with population of less than 50,000) against the growth of APCE in rural areas of individual states. There is indeed a positive relationship, but it is a relatively weak one.

Trends in the rural–urban dualism

An important issue in the development literature is the rural–urban gap in levels of income (and consumption) and in the incidence of poverty. Post-reform developments and globalization are sometimes viewed with concern as acting towards increasing the degree of this dualism.

Figure 2.2 Relationship between growths of APCE in small towns and rural areas.

Note

See full names of different states like WB, MP, AS, etc. in first column of Table 2.8.

We tried to throw some light on the variations in the gap between the incidences of poverty between rural and the urban areas across states. The variable to be explained is the ratio of the HCR in rural areas to that in the urban areas. Since the rural economy is large in all Indian states, a higher level of development in a state would generally imply a higher rural APCE. Also, with economic growth urbanization increases. A little reflection shows that the net impact of both these variables on the relative rural–urban gap depends on whether or not 'trickle down' is confined to the sector in question, or extends to the other sector. Take the expected sign of APCE (rural) first. If the impact is largely confined to the rural sector, then the relative gap would be reduced (the sign would be negative), but in so far as it reduces the level of the urban poor through higher demand for urban goods and services, the sign of the variable would be positive. The final outcome depends on the relative strength of the two forces. Similarly a higher rate of urbanization would mean a larger relative gap if urbanization has only a limited effect on rural incomes at the lower end of the scale, but would go the other way if the urban to rural linkage is strong. Second, as far as urban poverty is concerned, the size distribution of cities also matters since the incidence of poverty is inversely related to city size. We can use the summary measure of the share of metro population in the urban sector as a variable to capture this effect. The prediction about the sign of this variable, as with the other variables, depends on the relative strength of the linkage with the economies of sectors outside the metros. Generally, the incidence of poverty is lower in metro cities, so a greater preponderance of metro population would imply a higher relative gap in poverty incidence between the rural and the urban sector. But if 'trickle down' in states with a larger metro population is weak, the higher development of metro towns would have a limited effect on poverty incidence in non-metro urban areas, thus pushing up the HCR in the urban sector as a whole, i.e., the rural–urban ratio in HCR could be lower. The regression model with these variables is fitted to interstate variations in the rural–urban HCR ratio for different dates of the NSS rounds, and the results are reported in Table 2.7.

The results show that in the pre-reform years, between 1987–1988 and 1993–1994, the impact of the rate of urbanization is significantly negative: 'trickle down' extends to the rural economy. But the sign of the variable measuring the share of the metro in urban population is significantly positive: the growth of metro towns apparently reduces HCR in the urban sector as a whole, not so much in the rural areas. The sign of APCE (rural) is positive but not very significant.

In the post-reform years both the urbanization and metro share variables lose their significance, and the APCE is even less significant. Evidently factors other than those connected with the rural–urban 'trickle down' process discussed above now explained inter-state variations in the poverty gap. We have already seen that in the post-reform years urban growth was more important in reducing poverty across a wide range of states. This process reduced the inter-state variations in the relative poverty-gap even as it reduced the overall value of this gap.

We have also seen that within urban areas there was a distinction between the states which received a relatively large flow of FDI and those who did not. In the FDI states, it will be recalled, the pattern of poverty decline in urban areas by size of towns was different. It was inversely related to the size group of towns and was lowest in the largest cities. The relationship between HCR and town size was just the reverse in 'non-FDI' states. We tried to see if this difference was in any way related to the pattern of inter-state variability of the rural–urban poverty gap. The last equation of Table 2.7 shows that it is indeed so. FDI (urban) per capita is the only significant variable in the estimated equation and is positive, implying that those states which have a large FDI inflow have a significantly lower incidence of urban poverty relative to the rural sector. In other words, the effect of metro towns in increasing the relative poverty gap–which was true of the entire range of states in the pre-1993–1994 years–is now significant only in the FDI states. It draws attention to the point that FDI is an important player in the poverty scene in spite of the total inflow being much smaller in India than in other countries like China. FDI inflow reduces poverty significantly in the largest cities, but its 'trickle down' effect is limited outside the metro areas.

Differences by states

It is well known that Indian states differ substantially in the incidence of poverty. Also the distribution of population among the different states is uneven. The trend in the all-India measure of poverty, such as the HCR, will then be affected by the way the pattern of the difference in poverty reduction between more and less populous states. It has been hypothesized that growth rates and hence the rate of poverty reduction have not been generally stronger in the states with a larger share of the poor. The following paragraphs explore this hypothesis in more detail for the two periods we are considering.

Table 2.8 gives the shares of the individual states in the total count of those below the poverty line for the three dates (corresponding to the 38th, 50th and the 55th rounds of the NSS).

Seven states – Andhra Pradesh, Bihar, Madhya Pradesh, Maharashtra, Orissa, Uttar Pradesh and West Bengal–accounted for over 70 percent of the total poor in the rural areas in 1999–2000. Just three states–Bihar, Madhya Pradesh and Uttar Pradesh–accounted for over 40 percent of the rural poor. It is interesting to note that the same states to a large extent account for the bulk of the urban poor as well. The only difference between the two sets is Orissa which accounts for only 2.6 percent of the urban poor, reflecting the relative underdevelopment of the state. Looking back to 1987–1988 it is seen that there is not much difference in the spatial distribution of the poor–the same states account for the bulk of the rural and the urban poor. Perhaps the concentration of the poor in these states was a little higher in the earlier period.

It is interesting to see which states fell behind the all-India average in APCE growth rate in the most recent post-reform period. Because of its weight we look especially at the rural areas. The lagging states are: Assam, Andhra Pradesh, Bihar, Madhya Pradesh, Orissa, Rajasthan, Uttar Pradesh and West Bengal (for details see Mazumdar and Sarkar 2004, pp. 24–25). These states coincide with the set accounting for the bulk of the rural poor. The only state with large HCR missing in the set is Maharashtra–which is fairly close to the growth rate rural all-India. We conclude that the spatial pattern of growth rates in the 1993–2000 period was not favorable to the cause of poverty decline in the rural sector.

Poverty decline in the two periods

Further light can be shed on the experience of inter-state differences in poverty decline in the post-reform period by looking directly at the changes in the HCR. In Figures 2.5 through 2.8 we present the scatter diagrams of the state-level changes in the HCR over the two time periods between the NSS surveys – 1987–1988 to 1993–1994 and 1993–1994 to 1999–2000. We plot the change in the HCR on the y-axis against the level of HCR in the initial year of the respective period on the x-axis. The graphs are drawn separately for the rural and the urban areas.⁶

We would expect that the decline in HCR would be higher in the states where the absolute value of the HCR is higher. The percentage decline in HCR is given by the ratio of the two magnitudes. Convergence between states in the incidence of poverty will occur only if the percentage decline increases with the initial value of the HCR–i.e., the relationship between the two magnitudes in the scatter diagram is non-linear. The graphs (Figures 2.3 to 2.6), however, show that there is at best a linear relationship between the decline in HCR and its initial value. There is no evidence of inter-state differences in poverty incidence to converge over time in either sector.

Second, it is seen that while the slope of the line relating the initial HCR to its absolute change is more or less the same in the urban areas, it has definitely become flatter in the rural areas in the post-reform years. For the rural sector as a whole we can no longer say that the percentage decline in poverty is directly related to the absolute value of HCR in the rural areas of the state in the 1993–1994 to 1999–2000 period. The reason for this is that several of the states suffered from a deceleration of poverty decline in their rural sector in the post-reform areas. There is naturally an overlap between the states lying significantly below the regression line of Figure 2.6 and those identified earlier in this section as being laggards in rural APCE growth. They include Andhra Pradesh, Assam, Madhya Pradesh and West Bengal.

Table 2.9 showing the classification of states into four groups and their changes over the two periods might help to throw some further light on state-level changes in rural poverty decline. The groupings are from I to IV in descending order of poverty decline. Table 2.9 confirms that the flattening of the regression line in Figure 2.4 is largely due to four states slipping from Category I (high HCR decline) and category II (middling decline) in the pre-reform period to category IV in the post-reform one. This adverse effect is balanced to some extent by Maharashtra moving from IV to I, and Karnataka from IV to II.

Figure 2.3 Poverty (HCR) and declines in HCR from 1987–1988 to 1993–1994 (rural) across states.

Note

See full names of different states like WB, MP, AS, etc. in first column of Table 2.8.

Figure 2.4 Poverty (HCR) and declines in HCR from 1993–1994 to 1999–2000 (rural) across states.

Note

See full names of different states like WB, MP, AS, etc. in first column of Table 2.8.

Figure 2.5 Poverty (HCR) and declines in HCR from 1987–1988 to 1993–1994 (urban) across states.

Note

See full names of different states like WB, MP, AS, etc. in first column of Table 2.8.

Further details on the change in the HCR between the two periods for individual states is provided in Table 2.10 which reproduces the results of the decomposition analysis applied to the data for each state. The total percentage change in poverty for the state as a whole (rural and urban combined) is given in column 10, while the components of this statistic are found in the six columns preceding it. The column of %D gives the component due to differential growth, while %I is that due to change in inequality. The differences between HCR change and the sum of %D and %I in each sector are the sum of two elements: (i) 'Uniform growth'–the hypothetical growth if the APCE in each sector had grown at the same rate as in the state as a whole; (ii) the effect of the rural–urban shift (the latter has been small in most states). The last two columns of Table 2.10 give the percentage change in HCR in each of the two sectors – rural and urban separately.

Figure 2.6 Poverty (HCR) and declines in HCR from 1993–1994 to 1999–2000 (urban) across states.

Note

See full names of different states like WB, MP, AS, etc. in first column of Table 2.8.

Table 2.10 Decomposition of percentage change in head-count ratio (HCR)

We can use the table to classify the states into three groups in terms of the percentage change in HCR over the two periods we are considering (taking the rural and the urban sectors together):

1. states which accelerated the decline in HCR in the post-reform period markedly (more than doubled the percentage decline): Bihar, Gujarat, Maharashtra, Karanataka, Haryana, and Punjab;
   
   
2. states which had a less spectacular, but still substantial, decline in HCR relative to the earlier period: Tamil Nadu, Rajasthan, Uttar Pradesh and Kerala;
   
   
3. states which suffered retardation in HCR decline (or actually registered an increase in HCR) Andhra Pradesh, Madhya Pradesh, Orissa, West Bengal and Assam.

It is to be noted that the list in group 3 is identical to the group already mentioned earlier as the laggards in rural poverty decline. It shows the quantitative importance of the rural sector in the overall trend in poverty reduction.

While the growth of the rural sector is naturally the dominant influence on the overall decline in HCR (the share of the rural sector in the incidence of poverty being so much more), it is important to note that the urban sector became a much more important player in several states. We have already seen in section II that this is true for all-India in the post-reform years. But some individual states stand out. The urban sector of Gujarat, Karnataka and Punjab among the group 1 states, and Tamil Nadu and Rajasthan among the group 2 states contributed to the total HCR decline of to an extent of a third or more of the contribution of the rural sector. In all these states the differential rate of growth was higher in the urban sector, contributing to the HCR decline (%D in the urban sector was negative).

Of the states in which HCR declined Bihar and Uttar Pradesh are the only ones in which the rural sector increased its contribution to the decline. Rural growth also contributed strongly to HCR decline in Kerala, but it was offset to some extent by a substantial increase in inequality in the sector, so that the contribution of the urban sector increased in the 1993–1999 period. In the case of Bihar (and to some extent Uttar Pradesh, in so far its Eastern districts are really an extension of the economy of Bihar) the strong rural growth in APCE causing poverty decline is less due to the growth rate of its domestic rural economy, as it is to the remittance sent back home by migrant labor participating in the rural economy of the North-Western states and in urban areas scattered all over India.

We have already underlined in the third section the point about the increased inequality in the urban sector in the post-reform period for India as a whole. Its relative impact in retarding poverty decline is significant, but quantitatively not substantial. The state-level data reveal that the inequality effect has been more important in some states. Running down the column (9)–%I for the urban sector–it is seen that this has been so in Tamil Nadu, Gujarat, Haryana, Kerala and Punjab. It should not, however, be concluded that inequality increased only in the urban economy of some states. There was substantial inequality increase retarding poverty decline in the rural sector of a few states as well. These include Tamil Nadu, Kerala, Punjab, and to some extent Gujarat. It is interesting to note that these states are also the ones which experienced the adverse inequality effect in the urban sector as well.

It is important to comment on the experience of the group 3 states. Andhra Pradesh suffered from a slight retardation in the rate of decline in HCR, but the incidence of poverty actually increased in the other states in this group. It is seen that all these states suffered from adverse movements in the HCR in the rural sector. The figures of column (5) reveal that the crucial problem was the differentially lower growth in the rural economy which contributed to an increase in HCR. In the case of three of three states–Andhra Pradesh, Madhya Pradesh and West Bengal–the poverty incidence would have been worse but for the stronger performance of the urban sector. Assam stands out in registering an increase in inequality both in the rural and the urban sectors. An increase in urban inequality also contributed to HCR decline in Madhya Pradesh, but its relative importance was much less than that of the slow-down in rural growth.

Conclusions

In this chapter we have contrasted the change in poverty incidence in the 1987–1993 period with that in the 1993–1999 years. The second period could be seen as one in which the impact of the reforms tied to liberalization of the economy could be expected to have had some impact. We have based this comparison from several angles on the 'consumption expenditure survey' data generated by the NSS for three quinquennial rounds: the 43rd (1987–1988), the 50th (1993–1994) and the 55th (1999–2000). We have addressed the methodological problems involved in the surveys, and produced a new set of consistent data on mean consumption of the survey households, and the incidence of poverty. The differences between our estimates and those of other researchers have been spelled out in the first section and the Appendices.

We have a used a decomposition analysis of the percentage change in poverty over the two successive periods. This method used elsewhere by Mazumdar and Son (2002) enables us to quantify the relative contribution of three elements to the overall change in poverty incidence: mean growth of consumption, population shift between defined sectors and the change in inequality. The analysis is applied to the economy as a whole and to its rural and urban sectors. It is done for all-India and for the individual states.

The more important conclusions are the following:

1. At the all-India level the absolute change in the HCR was about the same in the post-reform period as in the previous one, but the rate of change was higher because the initial base was smaller. However, the growth rate of APCE increased rather more so that the elasticity of poverty change with respect to income fell. This is true even more for the poverty gap measure.
   
   
2. The share of the urban sector in poverty reduction increased in the second period. This was not due to a larger shift of population to urban areas–in fact the rate of this shift decreased. The major reason for the change was the higher differential growth rate of the urban sector. It was offset, but only partially, by an increase in urban inequality. However, the relative impact of rising urban inequality in retarding poverty decline in the late nineties was quantitatively not substantial.
   
   
3. Turning to poverty decline by size of towns there are two different aspects to the impact of the post-reform developments including globalization. On the one hand, there has been a distinct trend towards decentralization of economic activities to smaller towns and cities. This has led to the inverse relationship between growth and the size of towns observed in many states and in India as a whole. There are, however, a group of states in which the role of FDI is strong, and the impact is seen in a high growth rate in metro areas, so that a direct relationship between growth (and poverty reduction) and town size is observed, as was the case generally in the urban sector as whole in the previous period.
   
   
4. State-level analysis showed that the states could be divided into three groups when we compare the change in HCR in the post-reform years with the period preceding it:

group 1 are those which accelerated the decline in HCR in the post-reform period markedly (more than doubled the percentage decline): Bihar, Gujarat, Maharashtra, Karanataka, Haryana, Himachal Pradesh and Punjab;

group 2 states had a less spectacular, but still substantial, decline in HCR relative to the earlier period: Tamil Nadu, Rajasthan, Uttar Pradesh and Kerala;

group 3 states suffered retardation in HCR decline (or actually registered an increase in HCR): Andhra Pradesh, Madhya Pradesh, Orissa, West Bengal and Assam.

It is seen that the unhappy performance of the group 3 states is largely due to retardation in the rate of growth of the rural economy of these states. The growth of the urban sector, however, played a significant role in the poverty decline of group 1 and group 2. We have already seen that at the all-India level, urban growth in the post-reform years was higher than rural growth. Although the urban economy is still a small part of the economy, its contribution to poverty reduction started being important in several states. The urban sector of Gujarat, Karnataka and Punjab among the group 1 states, and Tamil Nadu and Rajasthan among the group 2 states contributed to the total HCR decline to as much as a third or more of the contribution of the rural economy.

5. Increased inequality seems to have been associated with the higher growth rate in the urban sector in the post-reform period for India as a whole. Its relative impact in retarding poverty decline is significant, but quantitatively not substantial. The state-level data reveal that the inequality effect has been more important in some states. They include Tamil Nadu, Gujarat, Haryana, Kerala and Punjab. It should not, however, be concluded that inequality increased only in the urban economy of some states. There was substantial inequality increase retarding poverty decline in the rural sector of a few states as well. Theses include Tamil Nadu, Kerala, Punjab, and to some extent Gujarat. It is interesting to note that these states are also the ones which experienced the adverse inequality effect in the urban sector as well.

Appendix 1

Poverty-decomposition methodology

Let us divide the total population into k mutually exclusive socioeconomic and demographic groups. For decomposable poverty measures, then, we can write the total poverty as the weighted average of poverty within each group.

where f_i and P_i are the population share and poverty index of the ith group, respectively. Further, define the change in poverty between two periods as

where and ., and P_2i being the poverty incidence

in the group in years 1 and 2, respectively, and f_1i and f_2i are the population shares of the ith group in years 1 and 2, respectively.

Equation (1) can be written as

which shows that the change in total poverty can be written as the sum of two components. The first component measures the effect on total change in poverty due to changes in within-group poverty and the second component estimates the change in total poverty due to possible shifts in population between groups.

The percentage change in total poverty, thus, can be written as follows:

where and .

Note that the first term in Equation (2) estimates the percentage change in total poverty explained by changes in poverty within groups. The second term estimates the percentage change in total poverty due to a shift in population between groups. The shift in population is deemed pro-poor if the second term is negative because it leads to a reduction in poverty. This situation is likely to occur if migration occurs from rural to urban areas. If migration takes place from urban to rural areas, on the other hand, the second component is likely to make a positive contribution to poverty. In this case, the population shift is not pro-poor.

Kakwani (2000) has proposed a decomposition, which explains the percentage change in poverty as a sum of two components: one is the growth effect, measuring the change in poverty when mean income changes but inequality remains fixed and the other component is the inequality effect, which measures changes in poverty when inequality changes but the mean income remains constant. This methodology can now be applied within each group.

A general poverty measure is characterized as

where z is the poverty line, μ is the mean income of society, and L(p) is the Lorenz curve. The Lorenz curve measures the effect of inequality on poverty. Following from Kakwani (2000), the percentage change in poverty can be written as

where (ΔP)_m is the change in poverty if mean income changes from μ₁ in period 1 to μ₂ in period 2 but the Lorenz curve remains fixed. Thus, (ΔP)_m can be written as

where L₁(p) and L₂(p) are the Lorenz curves in periods 1 and 2, respectively. Note that in deriving the mean effect, we can either fix the Lorenz curve for the initial period or for the terminal period. Because we do not know a priori which period of the Lorenz curve we should fix, we have taken the average of the two periods.⁷

Similarly, the inequality component can be derived as

which estimates the change in poverty if inequality measured by the Lorenz curve changes from L₁(p) in the initial period to L₂(p) in the terminal period but mean income is fixed between the two period. The sum of the mean and inequality effects gives rise to the total changes in poverty.

We apply the decomposition in (3) within each group, which results in

where

and

i^th group in year t (t = 1,2).

From (2) and (4), the percentage change in total poverty can be expressed as

The first term in equation (5) measures the effect of growth within each group on overall change in the poverty incidence, when the distribution within each group remains the same over time. This first term can be further decomposed into two terms:

where

and μ_2i = μ_1i(1 + g) where g being the average growth rate of the whole population is the mean income of the ith group in year 2 if the income of the ith group were growing at the same rate as the average growth rate of the whole population.

The first term in the right-hand side of (6) measures the effect of growth on percentage change in poverty under the counter-factual that all groups enjoyed the same uniform growth rates and the second term in the right-hand side measures the effect of differential growth rates within groups. Thus, substituting (6) into (5), we arrive at our poverty decomposition that expresses the percentage change in the poverty incidence as the sum of four components: (1) overall growth effect when inequality in the distribution does not change; (2) effect of differential growth rates in different groups; (3) effect of change in inequality within different groups; and (4) effect of changes in population shares between groups. This is an exact decomposition and, therefore, there will not be any residual term. This decomposition does not require us to specify an inequality measure. It uses the idea of shift in that part of the Lorenz curve, which affects the poor.

The first component will always be negative if there is a positive growth in the economy. The second component can be either negative or positive. If it is positive (negative), the disparity in growth rates of different groups has contributed to an increase (decrease) in total poverty. The third component can again be either positive or negative. If it is positive (negative), it indicates that a change in inequality within group has contributed to an increase (decrease) in the total poverty incidence. Finally, the fourth component measures the effect of migration of population between groups on the total poverty incidence.

Appendix 2

This appendix compares our calculation of HCR based on the adjustments made to the NSS figures on Average Per Capita expenditure (APCE) with others in the literature. We campared the APCE by fractile erxpenditure groups in rural and urban areas for the year 1993–1994 as given by Sundaram and Tendulkar 2003a (S&T) and our own calculation. It was observed that some differences exist only in highest fractile (95–100) but not for other lower fractiles. The effect on estimation on Level of HCR would be minimal.

In this regard, it is interesting to note that Sen and Himanshu (2004), following the S&T procedure, found all-India HCR for all areas to be 35.9 percent using URP and 30.6 percent using MRP for the year 1993–1994. S&T in their revised estimates found the corresponding figures to be 37.35 percent (URP) and 32.15 percent (MRP). This is probably because S&T use a somewhat different poverty line than the official poverty line. Sen and Himanshu (2004) further corrected the 55th round estimates of food and intoxicants for possible 'contamination' from the 7-day questionnaire. They used information from early NSS rounds to arrive at some estimates. At the lower bound, the extent of such contamination was found to be small but even then the authors calculated that the 55th round all-India poverty incidence using MRP was 27.8 percent as against the official figure of 26.1 percent. Thus they found the measured decline between 1993–1994 (MRP) and 1999–2000 (MRP revised) to be at most 2.8 percent implying an increase in the absolute number of the poor by five million. Their results are quite opposite to the official calculation of poverty decline (in HCR) by 9.8 percent implying a fall in the number of poor by 60 million and to the S&T revised estimates showing a fall of 4.83 percent denoting reduction in number of poor by 13 million in the period.

Document(s) 2 of 16