India occasional papers

India: Occasional Papers FILE COPY SWP279 World Bank Staff Working Paper No. 279 ,'- May 1978 The views and interpretations in this document are those of the authors and should not be attrkuuted to the World Bank, to its affiliated organizations or to any individual acting in their behalf. Prepared by: Montek S. Ahluwalia Development Research Center John Wall South Asia Programs Department Shlomo Reutlinger Development Economics Department Robert Cassen Consultant   The views and interpretations in this document are those of the authors and should not be attributed to the World Bank, to its affiliated organizations, or to any individual acting in their behalf. WORLD BANK Staff Working Paper No. 279 INDIA: OCCASIONAL PAPERS May 1978 Every year a great deal of economic work on India is undertaken by members of the Bank's regional and central staff and by consultants. The purpose of this volume is to bring certain of these papers to the attention of those who may be interested in these subjects. The Government of India has agreed to their being released in this way. Each paper appears under the name of its author, who is responsible for its content. The papers themselves span a gamut of subjects in which the Bank has shown interest in its economic work. Montek Ahluwalia's paper on rural poverty in India sifts from a set of confusing and conflicting time series evidence a view of the incidence of rural poverty over the past two decades. John Wall's paper on foodgrain management is a review of the elements and the performance of the Government of India's system to implement its policies with respect to the pricing, procurement, distribution, import and storage of grain. Shlomo Reutlinger's paper treats the subject of buffer stocking of foodgrain in a more rigorous analytical fashion, using a stochastic model designed to stimulate the salient features of Indian and world grain markets as they would affect the operation of a buffer stock. Robert Cassen's paper on human resources is a review of, first, India's basic demography, including a clarification of the vital statistics and population projections, and then of the family planning program's history and performance; the paper continues to discuss the population's nutrition and health and Government's programs to deal with these. Martin Wolf's paper on capital and growth records the striking phenomena of rising capital/output ratios in India during the last two decades, especially during the 1960's, and then decomposes these rises to see how they differ by major sectors. Prepared by: Montek S. Ahluwalia Robert Cassen Development Research Center Consultant John Wall South Asia Programs Department Martin Wolf Shlomo Reutlinger South Asia Programs Development Economics Department Department Copyright ' 1978 The World Bank 1818 H Street, N.W. Washington, D.C. 20433 U.S.A.   INDIA: OCCASIONAL PAPERS TABLE OF CONTENTS Page RURAL POVERTY IN INDIA: 1956/57 TO 1973/74 .... ......... 1 Montek S. Ahluwalia FOODGRAIN MANAGEMENT: PRICING, PROCUREMENT, 43 DISTRIBUTION, IMPORT AND STORAGE POLICY . John Wall THE LEVEL OF STABILITY OF INDIA'S FOODGRAIN CONSUMPTION . .................................... 92 Shlomo Reutlinger INDIA'S HUMAN RESOURCES ................................. 124 Robert Cassen CAPITAL AND GROWTH IN INDIA, 1950-71 ..................... 229 Martin Wolf   1 RURAL POVERTY IN INDIA: 1956/57 to 1973/74 Montek S. Ahluwalia October 1977   2 RURAL POVERTY IN INDIA: 1956/57 to 1973/74 TABLE OF CONTENTS Page PREFACE ............................................. 3 I. INTRODUCTION ......................................... 4 II. SOME PROBLEMS OF DATA AND METHODOLOGY .... ............ 5 A. Reliability of NSS Data ......................... 5 B. The Poverty Line in Current Prices .... .......... 6 C. Econometric Estimation of Poverty Measures ...... 8 III. TRENDS IN RURAL POVERTY 1956/57 TO 1973/74: NSS EVIDENCE ........ ............................... 13 A. All-India Results ............................... 13 B. Results for Individual States .... ............... 16 IV. ALTERNATIVE POVERTY ESTIMATES USING NATIONAL ACCOUNTS DATA ..................................... 20 V. AGRICULTURAL PERFORMANCE AND RURAL POVERTY .... ....... 24 A. The All-India Evidence .......................... 24 B. The Evidence for Individual States .... .......... 28 C. The Evidence on Relative Inequality .... ......... 35 VI. CONCLUSIONS ..................... ..................... 39 REFERENCES ..................... ...................... 41   3 RURAL POVERTY IN INDIA, 1956-57 to 1973-74 Preface The Green Revolution in wheat production in northwestern India had barely begun when some observers of Indian agriculture raised an alarm about the possibly adverse distributional impact of this new technology and the increased incomes that would flow from it. There was a widespread concern that only the economically and politically strong families in rural areas would benefit, that small farmers would not be able to adopt the new tech- nology and that traditional tenants would become landless as owners resumed self-cultivation. Larger farmers were seen to use their power to monopolize access to the needed inputs -- the new seeds, fertilizer, credit, water, and advice from the extension service -- all of which became scarce as the adop- tion rates rose. The new technology was also seen as raising the rent that would accrue to land suitable to irrigated wheat production and the return to the ability to manage the relatively complex requirements of the new technology. It was thought both of these forces led to the resumption of self cultivation and management by owners who had previously been content to let the land to tenants. This resumption was felt to result in the future impoverishment of lower income tenant families. It has proved very difficult to sift from a wide variety of un- substantiated assertions and fragmentary evidence a clear picture of what have been the trends in poverty and income distribution in rural areas of India. Many of those who have contributed to the controversy have had to rely on data that was severely limited in its usefulness for this purpose; one problem was in drawing conclusions based on comparisons between two isolated and arbitrarily chosen years; a serious limitation was that until recently suitable data was available only for the very first years of the Green Revolution period. This paper looks for trends and fluctuations about the trend in rural poverty by examining National Sample Survey data for fourteen years spanning 1956 to 1974. This evidence provides a fairly firm basis for generalizing about these issues. The major findings are that there has been no trend, up or down, in the incidence of absolute poverty over the period but there has been considerable fluctuation, with the incidence of poverty inversely related to agricultural performance. While this analysis is not the last word in controversy, it does tend to undermine the view sometimes encountered that agricultural progress through adoption of higher technology agriculture makes more difficult the alleviation of poverty.  I. INTRODUCTION* 1. Recent years have seen the development of an extensive and dis- quieting literature on trends in rural poverty in India and their relation- ship to agricultural growth. A recurring theme in much of this literature is that agricultural growth has been accompanied by a steady deterioration in distributional terms, involving not only an increase in relative in- equality but also an increase in absolute impoverishment. Indeed, it is argued that these trends are the natural consequence of the type of agri- cultural growth which can be expected within the existing institutional structure in Indian agriculture. 1/ This latter proposition has important implications for policy. It raises doubts about the scope for achieving even the fairly minimal welfare objective of alleviating absolute poverty in the future, at least through the kind of agricultural development that is currently deemed feasible, i.e. growth without radical institutional change. 2. The object of this paper is to evaluate the empirical basis for this assessment of past trends and future prospects. The principal data sources for our study are the various consumption surveys in urban and rural areas conducted by the National Sample Survey (NSS) which report the distribution of the population across per capita expenditure classes. These surveys have been used in several existing studies on rural poverty but these studies suffer from an important limitation. Most of them, for example Bardhan (1971), Rajaraman (1975) and Lal (1976), rely upon compari- sons between two arbitrarily chosen points in time. An early study by Minhas (1970) was based on data for seven years, but his time series extends only up to 1967/68. In this paper we will examine NSS data for 14 different years spanning the period 1956/57 to 1973/74. Only a systematic time series study of this type can provide a reliable basis for generalizing about trends in rural poverty. 3. Throughout this paper, our concern is principally with the extent of absolute poverty as distinct from the extent of inequality. Absolute poverty is defined with respect to a fixed poverty line in terms of real per * This paper was prepared for a Workshop on Analysis of Distributional Issues in Development Planning held in Bellagio, Italy in April 1977. I am indebted to Shail Jain for computer programming assistance and to P. Bardhan, B.S. Minhas, M. Lipton, T.N. Srinivasan and 0. Yenal for many helpful comments. 1/ For a sampling of the Indian debate on this question, see Bardhan (1971), Byres (1972), Dandekar and Rath (1971), Lal (1976), Raj (1976), and Rajaraman (1975). For a more general assertion of the operation of this process in a number of developing countries, see Griffin and Khan (eds.) (1976).  5 capita consumption. We have attempted, first, to document changes in inci- dence of poverty over time, and second, to relate these changes to two differ- ent measures of the extent or incidence of absolute poverty. The first measure is the percentage of the rural population below the fixed poverty line. The second measure is Sen's Poverty Index, which takes account not only of the percentage of the population in poverty but also of the gap between the poverty line. -and the- mean- consumpt-ion- of- -t-he- -poor-,- -as well- -as- t-he- ex-t-ent- of inequality amongst the poor. 1/ While the bulk of the debate has been conducted in terms of the percentage measure, the Sen Index has obvious advantages in measuring the true intensity of the poverty problem. 4. The paper is organized as follows. Section II deals briefly with some of the methodological problems encountered in our analysis. Section III presents our results on trends in the incidence of rural poverty over the period 1956/57 to 1973/74 for India as a whole as well as for the individual states. Section IV examines the results obtained by supplementing the NSS data with national accounts in making an assessment of trends in rural poverty. Section V attempts to relate observed changes in the incidence of poverty to some aggregate measures of agricultural performance. A summary view of the incidence on changes in rural poverty and factors affecting these changes is presented in Section VI. II. SOME PROBLEMS OF DATA AND METHODOLOGY 5. The use of NSS data to study changes in rural poverty over time raises a number of methodological problems. Three of these are particularly important for our purposes. First, there are obvious questions about the accuracy of NSS surveys in measuring the level and distribution of consumption. Secondly, there are problems associated with defining consumption expenditure levels in current prices of different years, which correspond to some constant "poverty line." Finally, there are estimation problems related to the methods adopted for estimating our measures of the incidence of poverty, given the NSS consumption distribution and a poverty line in consumption expenditure. Each of these problems, and our particular approach to resolving them, is discussed briefly below. (A) Reliability of NSS Data 6. How reliable are the NSS estimates of the level and distribution of consumption and are they comparable over time? One approach to answering this question is to examine the design and implementation of NSS surveys and to reach a judgment on their reliability on this basis. A second approach is to compare NSS estimates of consumption with other estimates. Since the two approaches are not exclusive of each other, it is instructive to follow both. 7. In terms of design and implementation, there is substantial agree- ment that the NSS data provide valuable documentation of the level and dis- tribution of consumption expenditure (in current prices) and changes in these 1/ See Sen (1974).  6 magnitudes over time over the past twenty years. The surveys are based on scientifically selected (stratified random) samples and are conducted by a permanent staff of trained investigators. The use of interpenetrating sub- samples permits estimates of sample error for magnitudes whose sampling distribution is not known and the sample variance for such estimates has been shown to be quite small. The concepts and definitions used have been sufficiently standardized from the mid-fifties onwards for the results to be comparable over time. 1/ Our study is based on the results of fourteen surveys conducted between 1956/57 and 1973/74. In all cases, the reference period over which consumption data were collected was one month. Both con- sumption out of home grown stocks and purchased consumption are included in the valuation of consumption, with the former being valued at ex-farm prices and the latter at retail prices. 2/ These features clearly suggest a high degree of reliability for estimates at a point in time as well as comparabil- ity of these estimates over time. 8. Although there is general agreement on the high quality of NSS surveys, it has been argued that NSS estimates of consumption (i) understate per capita consumption levels in the economy as a whole in comparison with the official private consumption estimates contained in the National Accounts, and (ii) the degree of understatement increases sharply after the mid-sixties. This discrepancy between the NSS and the National Accounts has been exten- sively discussed in the Indian literature and there is no easy way of resolv- ing the problem. Since the choice between these alternative sources of in- formation has important implications for our estimates of trends in poverty, we have examined both data sources. While the bulk of the paper focusses on the NSS evidence, the results obtained from using National Accounts data as a supplementary source are discussed in Section IV. (B) The Poverty Line in Current Prices 9. The determination of "equivalent" poverty lines in terms of consump- tion expenditure for different years (in current prices) corresponding to some constant poverty level raises formidable problems. The conventional approach is to proceed in two steps. First, a poverty line is defined for a base year in terms of the level of consumer expenditure required to achieve some speci- fied consumption bundle at base year prices. Equivalent consumption levels (in current prices) are then calculated for other years by using a "suitable" price index. There are difficulties associated with each step. 1/ The earlier surveys experimented with weeks and months as alternative reference periods until the 7th Round (Oct. 1953 - Mar. 1954) onwards when the reference period was standardized to one month. The imputation method used to value homegrown produce was changed from the 9th Round (May 1955 - Nov. 1955) onwards, with consumption out of homegrown stock being valued at ex-farm prices, instead of at retail prices as in previous Rounds. 2/ For the purpose of valuation, purchased consumption includes both cash purchases and purchases through exchange of commodities or labor services. Thus consumption out of wages in kind is valued at retail prices.  7 10. The basic problem with our first step - the definition of a poverty line - is that any such line is necessarily arbitrary. In this paper, we finesse the problem by choosing our poverty line primarily to conform with past practice, without trying to justify it in terms of some independently measurable real living standard. Accordingly, the poverty line used through- out this paper is a consumer expenditure level of Rs 15 per person for 30 days at 1960/61 rural prices. This line has an established pedigree in the Indian literature. As shown by Bardhan (1971), an expenditure level of Rs 15 in 1960/61 at rural prices corresponds to Rs 20 per person at all India 1960/61 prices, which is the minimum level originally adopted by the Planning Commission in 1962 and subsequently referred to in the Approach to the Fifth Plan. 1/ We note that Dandekar and Rath (1970) also adopted this line .for measuring rural poverty on the grounds that it corresponded to the expenditure level at which food consumption (on average) provided the "norm" of 2,250 calories per day. However, it is important to emphasize that attempts to interpret this line as corresponding to a nutritional minimum could be seriously misleading. 2/ Suffice it to say that this level of expenditure represents an extremely low level of living and one that has been accepted as a "minimum level" in the policy debate. 11. Since the NSS data on consumption distribution are available separately for each state, it is possible to estimate the incidence of poverty separately for each state. This analysis calls for state specific poverty lines. Commodity prices vary significantly across states so that the same real consumption level requires different levels of nominal con- sumption. Furthermore, rates of inflation across states are also different. Bardhan (1971) has estimated rural poverty lines for each state in 1960/61 corresponding to all-India rural poverty line of Rs 15. We have adopted these estimates as our base year estimates of the poverty line in each state. 12. The second step in estimating poverty lines for different years is the identification of a suitable price index for the rural poor and this presents another problem. Consumer price indices for the "average" consumer are clearly not suitable since the budget of the poor is much more heavily weighted towards those items which have displayed very high inflation rates in the sixties (e.g. food and especially coarse grains). This problem has been extensively discussed in the Indian literature and there is general agreement on the need for separate price indices for the various fractiles of consumer expenditure reflecting differences in the commodity composition 1/ See Planning Commission (1962). 2/ Not only is it extremely difficult to measure the actual calorie intake from the consumer expenditure level as recorded in the NSS, it is also arguable that the use of 2,250 calories as a minimum requirement is inappropriate. For a discussion of the difficulties in translating the notion of a nutritional minimum into a poverty line in terms of consump- tion expenditure, see Sukhatme (1977). See also Bliss and Stern (1975).  8 of consumption. Although some attempts have been made in this direction by researchers such price indices are not yet available. 1/ 13. In the absence of price indices specially designed for the rural poor, we have followed Bardhan (1971) in using the Consumer Price Index for Agricultural Laborers (CPIAL) prepared by the Labor Bureau as the most appropriate index for our purpose. These indices are available for India as a whole and separately for each state and are presented in Table 1(a). 2/ It is worth noting that the CPIAL for India as a whole shows a substantially higher inflation rate than the implicit deflator for consumption in the national accounts. Given the NSS estimates of consumption in current prices, the use of the CPIAL rather than the national accounts deflator for consumption will lead to larger estimates of the incidence of poverty. 14. Applying the CPIAL to the base year estimates of the poverty line, in terms of consumption expenditure per person in 1960/61 prices, we can calculate equivalent poverty lines for each of the years for which NSS con- sumption distributions are available.. The resulting estimates of the poverty line for each state, and for India as a whole, for fourteen years spanning the period from the late fifties to the early seventies are shown in Table 1(b). These estimates are obviously subject to all the limitations arising from the use of the CPIAL as the price index. More seriously, it can be argued that the very approach of using a base weighted price index is flawed since it cannot reflect the impact of changing relative prices upon the commodity composition of consumption. 3/ In defense of our estimates we can only state that with all these limitations, they are probably the best estimates avail- able, and are certainly in line with past practice. (C) Econometric Estimation of Poverty Measures 15. Given the NSS consumption distribution and the poverty line, we need to estimate the two poverty measures used in this paper, i.e. the per- centage of the population below the poverty line and Sen's Poverty Index. The method used to estimate each of these measures is described below. 1/ For an attempt to construct fractile specific price indices for the rural population in the sixties see Vaidyanathan (1974). Vaidyanathan finds that the price index for the poor rose slightly faster than for the non- poor. 2/ The footnotes to Table 1(a) give details on supplementary data sources used and several assumptions made to obtain estimates for missing years and missing states. 3/ This problem can only be resolved if the poverty line in current prices is determined for each year separately as the solution to the program- ming problem of defining the minimum cost budget needed to achieve some specified minimum requirements (in nutritional and other terms) subject to any desired set of taste and preference constraints. Needless to say, no such exercise yielding poverty lines over time has been conducted for India or for any of the states.  Table 1(a) CONSUMER PRICE INDICES FOR AGRICULTURAL LABOUR 1956-57 1957-58-/ 1959-60 1960-61 1961-62 1963-641/ 1964-65 1965-66 1966-67 1967-68 1968-69 1970-71 1973-74 Andhra Pradesh 91 91.1 99 100 101 104.6 125 137 157 161 167 171 242 Assam 92 104.5 97 100 99 113.0 131 145 190 219 208 203 263 Bihar 99 108,4 102 100 103 118.0 150 179 229 250 187 206 337 Gujarat 98 98.5 94 100 102 105.8 133 136 153 160 160 173 246 Karnataka 89 88.8 99 100 100 109.5 146 171 179 187 182 188 275 Kerala 99 94.6 101 100 106 109.1 132 150 161 172 195 214 276 Madhva Pradesh 101 104.7 101 .100 103 119.5 139 156 202 219 195 198 309 Maharashtra 98 98.5 103 100 98 110.9 151 159 175 133 177 192 276 Orissa 95 6.8 99 100 102 131.0 143 162 191 209 218 212 282 Punjab and Haryana 99 100.5 103 100 104 114.3 139 138 174 193 193 194 273 Rajasthan 96 91.8 n.a. 100 94 103.7 132 140 167 173 182 173 284 Tamil Nadu 98 98.5 105 100 113 121.0** 138 143 175 172 177 174 242 Uttar Pradesh 101 107.5 103 100 100 133.0** 164 165 211 236 179 183 303 West Bengal 100 107.9 110 100 105 132.8 136 140 167 241 200 206 276 ALL INDIA' 97.2 100.9 101.7 100 103 118 142 155 190 206 185 192 286 Implicit Consumption n.a. n.-. n.a. 100 104.3 116.5 127.1 137.6 156.1 176.5 169.8 179.8 241.0 1 Deflator in National Accounts Sources: Price indices for individual states and for India as a whole relate to the agricultural year July-June and are prepared by the Bureau Labour Bureau and reported in the Indian Labour Journal. However, these do not include estimates for some of the states (Tamil Nadu, Uttar Pradesh and Rajasthan) for years preceding 1964-65. For these states we have used estimates prepared by other researchers. We have used A.V. Jose (1974) as the basic source supplemented by Lal (1976). 1/ Calculated from Indian Labour Journal on assumption that index in July-August 1963 was the same as the average for September-December 1963. Figures for Tamil Nadu and Uttar Pradesh are averages of 1962-63 and 1964-65 obtained from Lal (1976). 2/ Calculated by averaging calendar years 1957 and 1958 from Indian Labour Journal. Figures for Gujarat, Maharashtra, Tamil Nadu and Uttar Pradesh are obtained from Lal (1976) as averages of 1956-57 and 1958-59.  Table l(b) RURAL POVERTY LINES: CONSUMPTION l'ER PERSON FOR 30 DAYS (Rs. in current prices) 1956-57 1957-58-/ 1959-60 1960-61 1961-62 1963-64-/ 1964-65 1965-66 1966-67 1967-68 1968-69 1970-71 1973-74 Andhra Pradesh 14.1 14.1 15.4 15.5 15.7 16.2 19.38 21.24 24.34 25.0 25.9 26.5 37.5 Assam 15.0 17.0 15.8 16.3 16.1 18.4 21.4 23.64 30.97 35.7 33.9 33.1 42.9 Bihar 15.6 17.1 16.1 15.8 16.3 18.6 23.7 28.28 36.18 39.5 29.5 32.5 53.2 Gujarat 16.5 16.5 15.8 16.8 17.1 17.8 22.3 22.85 25.70 26.9 26.9 29.1 41.3 Karnataka 13.9 13.9 15.4 15.6 15.6 17.1 22.8 26.68 27.92 29.2 28.4 29.3 42.9 Kerala 15.9 15.2 16.3 16.1 17.1 17.6 21.3 24.15 25.92 27.7 31.4 34.5 44.4 Madhya Pradesh 14.2 14.8 14.2 14.1 14.5 16.8 19.6 22.0 28.48 30.9 27.5 27.9 43.6 Maharashtra 15.7 15.8 16.5 16.0 15.7 17.7 24.2 25.44 28.0 29.3 28.3 30.7 44.2 Orissa 13.8 14.0 14.4 14.5 14.8 19.0 20.7 23.49 27.70 30.3 31.6 30.7 40.9 Punjab and Haryana 15.7 16.0 16.4 15.9 16.5 18.2 22.10 21.94 27.67 30.7 30.7 30.8 43.4 Rajasthan 14.1 13.5 - 14.7 13.8 15.2 19.4 20.58 24.55 25.43 26.8 25.4 41.7 Tamil Nadu 16.1 16.1 17.2 16.4 18.5 19.8 22.6 23.45 28.7 28.2 29.0 28.5 39.7 Uttar Pradesh 14.6 15.6 14.9 14.5 15.2 19.3 23.8 23.93 30.60 34.22 26.0 26.5 43.9 West Bengal 18.1 19.5 19.9 18.1 19.0 24.0 24.6 25.34 30.23 43.6 36.0 37.3 50.0 All India 14.58 15.1 15.3 15.0 15.45 17.7 21.3 23.25 28.50 30.9 27.8 28.8 42.9  11 16. Estimating the percentage of the population in poverty poses prob- lems because our primary data source is the distribution of the rural popu- lation (i.e. persons) by discrete per capita household expenditure classes. This permits us to read from the data the cumulative percentage of the pop- ulation lying below an upper or lower end of an expenditure class. However, the poverty line typically will not coincide with a boundary between expen- diture classes. This means that either we use some method to characterize the distribution continuously within expenditure classes, or we must content ourselves with reporting a range for the incidence of poverty, the limits -being given by the cumulative percentage of population up to the upper and lower ends of the particular expenditure class within which the poverty line falls. Thus Bardhan (1971) reported range estimates for the incidence of poverty by state in 1960/61 and 1967/68. This method is unsatisfactory for our purposes since the range is typically quite wide (over ten percentage points of population in most cases) making it difficult to generalize about time trends in the estimated incidence of poverty. 17. The approach adopted in this paper is to characterize the distri- bution continuously in terms of the Lorenz curve of consumption and use the continuous Lorenz curve to estimate poverty measures. We have used the NSS data to estimate a Lorenz curve of consumption for each year, using the method proposed by Kakwani and Podder (1976). This consists of defining a new coordinate system for the Lorenz curve (see Fig. 1) by expressing n , the perpendicular distance between Lorenz curve and the line of perfect equality, as a function of Tr , the distance between the origin and the point Q on the line of perfect equality at which it is intersected by a perpendicular from the Lorenz curve (see Fig. 1). Kakwani and Podder use the following equation for the Lorenz n afl ff Note that v2 is the length of the line of perfect equality given that it is the diagonal of the unit square. Clearly, T = 0 when 1= 0 and also when v = f2 , i.e., the Lorenz curve meets the line of perfect equality at the two ends of the unit square. This equation can be estimated in log linear form using ordinary least squares. 18. The estimated parameters of the fitted Lorenz curve can then be used to obtain a point estimate of the percentage of the population below a given poverty line. This procedure can be illustrated with reference to Figure 1, which shows the Lorenze curve for the distribution of consumption. The slope of the Lorenz curve at the point x is equal to the ratio of the consumption level of the person at the top of the lower xI percent of the population to the mean consumption of the whole population. Our procedure amounts to using the parameters of the estimated Lorenz curve to calculate the point x at which the slope of the curve equals the ratio of the poverty line to the mean consumption level, and then reading the cumulative percentage of population (x ) corresponding to the point x on the Lorenz curve.  12 19. The estimated Lorenz curve can also be used to compute Sen's Poverty Index. The Index is defined as follows: I x Ic* - c tl-G ) c* p p p p where x = percentage of population below poverty line c* = poverty line p c = mean consumption of the poor G = Gini coefficient of consumption of the poor We have seen above how x can be calculated from the estimated Lorenz curve. The mean consumption of the poor c is simply (y . x) . c where y is the Cumulative 1 Proportion of Income Cumulative Proportion of Population 0 1 Figure 1 share of consumption by the_poor in total consumption (which can be estimated from the Lorenz curve) and c is the mean consumption of the population as a whole as recorded in the NSS. The Gini coefficient of the distribution of consumption among the poor can be computed once the Lorenz curve is known. It is easily seen that the Sen Index ranges from 0 to 1.  13 III. TRENDS IN RURAL POVERTY 1956/57 TO 1973/74: NSS EVIDENCE 20. We now turn to the evidence on trends in poverty in rural India, based on the data and methodology discussed above. The poverty lines pre- sented in Table l(a) have been used to estimate the percentage of the rural population below this level of per capita consumption, as well as Sen's Poverty Index, for various years for India as a whole, and for individual states. The results are discussed separately below. (A) All-India Results 21. Our estimates of the two poverty measures for rural India as a whole are presented in Table 2. The NSS data permit two different estimates of the percentage of the rural population in poverty in India as a whole. Estimate I is obtained by applying the all-India poverty line for various years (see Table la) to the NSS consumption distribution for rural India. Estimate II is obtained as a weighted sum of the estimated percentages in poverty in individual states, obtained from the NSS distributions for indi- vidual states and the state specific poverty line. As there were substantial interstate differences in prices in the base year, and furthermore, inflation occurred at different rates across states, it can be argued that Estimate II, which is based on state specific poverty lines, is a better estimate of the percentage of the rural population in poverty. 22. The most important feature of the results presented in Table 2 is the marked fluctuation over time in the extent or incidence of rural poverty. The percentage in poverty declines initially from over 50 percent in the mid- fifties to around 40 percent in 1960/61, rises sharply through the mid-sixties, reaching a peak in 1967/68, and then declines again. The Sen Index also dis- plays the same pattern. Since this index reflects not only the percentage below the poverty line, but also the average shortfall of this group from the poverty line, it is reasonable to conclude from the range of variation in this index that we are measuring substantial fluctuations in the intensity of poverty and not merely marginal shifts of large numbers from a position slightly above the poverty line to a position slightly below. 23. It is important to determine whether the observed fluctuations arise solely from the sampling variation over time in our estimates, or whether they reflect genuine changes in the incidence of poverty arising from underlying economic factors. Fortunately, NSS surveys are conducted on the basis of interpenetrating sub-samples and differences between estimates of the incidence of poverty based on different sub-samples for the same year provide an indica- tion of the range of variation in sample estimates. Separate tabulations of the data by sub-sample are available for the earlier years (up to 1965/66) and have been used to obtain sub-sample estimates of each of our poverty measures for India as a whole (see figures in parentheses in Table 2). The range of variation between sub-sample estimates for the same year is clearly much smaller than the variations observed over time. This suggests that the measured fluctuation in the incidence of poverty reflects real changes in the severity of the poverty problem over time.  14 24. The existence of fluctuations over time implies that we cannot generalize about underlying trends on the basis of comparisons between selected endpoints. For example, Bardhan (1971) reported a sharp increase in the incidence of rural poverty between 1960/61 and 1968/69 while Lal (1976) has argued that the incidence of rural poverty declined between 1956/57 and 1970/71. 1/ Our estimates for these years (derived by a different estimation method) confirm the direction of change between these particular endpoints as reported in each study, but they also point to the danger of using such comparisons for any assessment of underlying trends. This can only be done on the basis of the time series as a whole. 25. A linear time trend fitted to each of our two estimates of the per- centage in poverty and our estimates of the Sen Index yields the following results (terms in parentheses are t ratios). 2 Estimate I - 45.59 + 0.393 T R = .09 (13.78) (1.04) 2 Estimate II = 47.74 + 0.262 T R = .06 (14.10) (0.81) Sen Index = 0.185 + 0.001 T R- .01 (8.95) (0.38) These results provide no evidence for asserting a trend increase or decrease in rural poverty over the period as a whole. The increasing incidence of poverty over the sixties reported by Bardhan appears as an upswing in a pattern of cyclical variation. We note that our conclusion also differs from Minhas (1970), who reported a decline in the percentage of the rural population in poverty over the period 1956/57 to 1967/68. This difference is due to Minhas' use of national accounts data to measure the level and growth of rural consumption over time. This approach is examined in detail in Section IV. 26. The absence of any discernible trend in the incidence of poverty obviously implies an increase in the absolute numbers of people in poverty because of the growth in rural population over the period. The absolute size of the poor population also fluctuates over the period, as shown in Table 2, but in this case, fitting a time trend to each estimate (P and P2 respectively) yields the following results: 1 1/ Lal's findings are based on data for five states only, and furthermore, on the incidence of poverty within the so-called weaker sections of the population. This comparison is somewhat questionable for the purpose of determining changes in the incidence of rural poverty since it is possible that the incidence of poverty in the weaker sections may decline while the incidence of poverty in the rural population as a whole may increase. This can happen if the proportion of the rural population in the "weaker sections" category rises sufficiently over the period.  15 Table 2 NSS BASED ESTIMATES OF RURAL POVERTY IN INDIA SIZE OF POVERTY POPULATION PERCENTAGE OF RURAL POPULATION IN POVERTY SEN'S POVERTY INDEX A/ (millions) Estimate I Estimate II Based on Based on Derived Derived All India State Specific from from Poverty Line Poverty Line Estimate I Estimate II 1956-57 54.1 n.a. .228 181.0 n.a. (Subsample 1) (53.5) (.222) (Subsample 2) (54.7) (.235) 1957-58 50.2 53.4 .220 1;71.4 182.4 (Subsample 1) (48.6) (.210) (Subsample 2) (51.7) (.229) 1958-59 46.5 n.a. .193 161.8 n.a. (Subsample 1) (47.9) (.198) kSubsample 2) (44.9) (.186) 1959-60 44.4 48.7 .168 15,7.6 172.8 (Subsample 1) (46.3) (.175) (Subsample 2) (42.5) (.161) 1960-61 38.9 42.0 .143 141.2 152.4 (Subsample 1) (38.4) (.141) (Subsample 2) (39.3) (.148) 1961-62 39.4 42.3 .136 145.9 156.6 (Subsample 1) (41.5) (.148) (Subsample 2) (39.8) (.136) (Subsample 3) (36.9) (.134) 1963-64 44.5 49.1 .159 171.4 189.1 (Subsample 1) (46.0) (.168) (Subsample 2) (45.7) (.170) 1964-65 46.8 50.4 .174 183.8 197.9 (Subsample 1) (46.8) (.175) (Subsample 2) (46.8) (.173) 1965-66 53.9 51.1 .214 215.9 204.7 (Subsample 1) (54.7) (.220) (Subsample 2) (53.0) (.210) 1966-67 56.6 57.4 .242 231.2 234.5 1967-68 56.5 57.9 .237 235.3 241.1 1968-69 51.0 53.5 .201 216.6 227.2 1970-71 47.5 49.1 .176 209.5 216.6 1973-74 n.a. 47.6 n.a. n.a. 221.3 1/ As explained in the text, thls index ranges frcn 0 to 1.  16 2 P = 145.50 + 5.333 T R = .55 1 (12.83) (3.6) 2 P = 153.99 + 4.855 T R = .60 2 (11.67) (3.84) In other words, the absolute number of the rural poor has grown significantly over time, increas-ing--or-a-v-e-rge-b-y--abbut5 5million every year. (B) Results for Individual States 27. Our estimates of the two poverty measures for individual states are reported in Tables 3(a) and 3(b) respectively. In general, the time pattern of the incidence of poverty in individual states follows the pattern of fluctuation described for India as a whole. The incidence of poverty for almost all states, using either of the two poverty measures, declines up to the early sixties and then begins to rise again, reaching a peak in 1967/68 or 1968/69, and declining again thereafter. 28. Once again, we have tested for the existence of an underlying trend by fitting a linear time trend to each of our poverty measures for individual states. The results are summarized in Table 4. Only two states (Assam and West Bengal) show a significant trend increase in poverty, according to both our poverty measures. Andhra Pradesh and Tamil Nadu show a trend decline in the incidence of poverty although in Tamil Nadu this is significant only when we use the Sen Index. For all other states there is no significant time trend in the incidence of rural poverty using either measure. 29. Our results for Punjab and Haryana are particularly worth noting in view of the findings of other studies that the incidence of rural poverty has increased in this region, despite the visible success of the Green Revolution in raising output. For example, Bardhan (1971) found that the incidence of poverty in Punjab and Haryana increased between 1960/61 and 1967/68 and Rajaraman (1976) reported that the incidence of poverty in Punjab (excluding Haryana) increased between 1960/61 and 1970/71. We find that while both conclusions stand as long as we compare these particular endpoints, they are misleading in respect of underlying trends. The incidence of rural poverty in Punjab and Haryana appears to have been unusually low in 1960/61 and estimates for this year therefore should not be used as the base for comparison. 1/ Consideration of the time series as a whole clearly does not suggest a trend increase in the incidence of poverty. 1/ One possible reason for this is the small size of the sample. The sample size for Punjab and Haryana in 1960/61 was only 142 households compared with 243 in 1959/60 and 224 in 1961/62.  Table 3(a) PERCENTAGE OF RURAL POPULATION IN POVERTY BY STATES 1957-58 1959-60 1960-61 1961-62 1963-64 1964-65 1965-66 1966-67 1967-68 1968-69 1970-71 1973-74 Andhra Pradesh 53.5 48.8 50.1 47.2 45.6 41.5 45.4 47.9 46.0 47.3 41.0 39.8 Assam 28.0 31.4 25.6 29.4 24.4 24.2 31.3 46.8 38.4 47.3 35.3 39.3 Bihar 59.7 55.7 41.5 49.9 52.3 54.3 59.4 74.4 70.9 59.4 59.0 58.4 Gujarat 1/ 41.5 31.6 39.7 45.7 49.8 50.7 54.1 50.8 42.8 43.8 35.6 Karnataka 41.3 48.9 39.1 35.4 50.5 55.1 63.9 59.5 56.9 58.8 47.2 46.9 Kerala 59.6 62.3 57.8 50.3 52.8 60.7 70.7 67.1 63.4 64.6 62.0 49.3 Madhya Pradesh 57.7 46.4 43.8 40.0 43.6 42.1 47.2 58.3 62.3 56.0 52.9 52.3 Maharashtra 1/ 54.5 48.4 43.6 48.2 59.1 57.8 63.2 57.2 54.8 46.6 49.8 - Orissa 66.6 63.4 62.4 49.3 60.0 61.9 62.1 64.2 64.7 71.2 65.0 58.0 Punjab & Haryana-/ 28.0 24.2 18.8 22.3 29.4 26.5 26.5 29.5 33.9 24.0 23.6 23.0 Rajasthan 33.4 n.a. 32.3 33.0 32.6 31.8 30.8 37.1- 35.9 41.4 41.8 29.8 Tamil Nadu 67.8 64.4 53.9 51.0 52.0 57.4 59.5 62.7 58.1 60.6 57.3 48.3 Uttar Pradesh 52.3 36.7 37.9 35.4 56.6 53.7 47.1 55.2 60.2 46.4 40.6 47.3 West Bengal 62.3 61.4 40.4 58.3 63.3 64.0 56.5 64.3 80.3 74.9 70.1 66.0 INDIA Estimate II 53.4 48.7 42.0 42.3 49.1 50.4 51.1 57.4 57.9 53.5 49.1 47.6 (Weighted Averages) 1/ Figures for Gujarat and Maharashtra are not available separately for the year 1957-58 since NSS tabulations for that year refer to the old Bombay State including both Maharashtra and Gujarat. The poverty incidence for Bombay State is 56.2 and this figure has been used with the combined weights for Gujarat and Maharashtra to calculate the all-India weighted average. 2/ NSS data report a single distribution for the old Punjab State (including Haryana) up to 1963-64, after which separate distributions are reported for Punjab and Haryana. The poverty incidence for the years after 1963-64 is based on a pooling of the data for the two states, using rural populations of Punjab and Haryana as recorded in the 1971 Census as weights. It should be noted, however, that parts of the old Punjab States were merged into Himachal Pradesh and the Union Territory of Delhi. Our procedure ignores this problem but the error is likely to be extremely small.  Table 3(b) SEN'S POVERTY INDEX FOR INDIVIDUAL STATES 1957-58 1959-60 1960-61 1961-62 1963-64 1964-65 1965-66 1966-67 1967-68 1968-69 1970-71 1973-74 Andhra Pradesh 0.22 0.17 0.18 0.18 0.16 0.14 0.16 0.18 0.17 0.18 0.13 0.14 Assam 0.07 0.08 0.05 0.06 0.06 0.05 0.08 0.13 0.10 0.14 0.08 0.11 Bihar 0.29 0.23 0.15 0.19 0.21 0.22 0.25 0.41 0.37 0.26 0.25 0.24 Gujarat 0.15 0.12 0.12 0.16 0.17 0.20 0.22 0.20 0.16 0.15 0.10 Karnataka 0.17 0.19 0.13 0.12 0.18 0.21 0.30 0.27 0.24 0.25 0.17 0.17 Kerala 0.29 0.29 0.25 0.21 0.21 0.29 0.33 0.31 0.28 0.31 0.29 0.20 Madhya Pradesh 0.27 0.19 0.17 0.14 0.17 0.15 0.17 0.26 0.29 0.24 0.21 0.20 Maharashtra 0.21 0.18 0.15 0.17 0.24 0.22 0.27 0.22 0.20 0.16 0.18 Orissa 0.32 0.28 0.31 0.20 0.26 0.26 0.26 0.27 0.28 0.33 0.30 0.23 Punjab & Haryana 0.10 0.07 0.05 0.07 0.10 0.08 0.08 0.08 0.11 0.05 0.07 0.06 Rajasthan 0.14 n.a. 0.10 0.12 0.11 0.11 0.10 0.14 0.14 0.17 0.16 0.09 Tamil Nadu 0.34 0.30 0.24 0.22 0.21 0.23 0.25 0.26 0.24 0.25 0.22 0.17 Uttar Pradesh 0.22 0.13 0.14 0.12 0.23 0.21 0.17 0.23 0.25 0.17 0.13 0.15 West Bengal 0.26 0.25 0.14 0.20 0.26 0.26 0.22 0.27 0.40 0.33 0.31 0.31  19 Table 4 LINEAR TIME TREND FOR POVERTY INCIDENCE BY STATE Percentage of Rural Population Sen's Poverty Index in Poverty as Dependent Variable as Dependent Variable 2 Time 2 Constant Time R Constant (x10-2) R Andhra Pradesh 52.50 -0.671** .64 .204 -.38** .56 (31.84) (4.25) (18.38) (3.58) Assam 23.47 1.060** .38 .049 .38** .35 (5.29) (2.50) (2.85) (2.30) Bihar 50.83 0.752 .17 .214 .44 .08 (9.15) (1.41) (4.38) (0.95) Oujarat 42.18 0.199 .02 .158 .02 .00 (7.27) (0.37) (5.00) (0.05) Karnataka 42.76 0.800 .18 .166 .35 .10 (7.65) (1.49) (4.66) (1.03) Kerala 59.84 0.022 .00 .276 -.04 .00 (13.04) (0.05) (9.10) (0.14) Madhya Pradesh 44.95 0.559 .13 .186 .20 .04 (9.39) (1.22) (5.57) (0.64) Maharashtra 51.85 0.116 .01 .198 .03 .00 (10.12) (0.25) (6.29) (0.11) Orissa 61.30 0.117 .01 .284 -.10 .01 (16.64) (0.33) (10.88) (0.39) Punjab & Haryana 25.26 0.058 .00 .084 -.07 .03 (8.997) (0.22) (6.17) (0.56) Rajasthan 30.91 0.300 .08 .1163 .08 .02 (8.120) (0.86) (5.62) (0.39) Tamil Nadu 62.59 -0.514 .18 .303 -.62** .46 (17.15) (1.47) (13.66) (2.93) Uttar Pradesh 43.92 0.375 .04 .179 .01 .00 (7.68) (0.68) (5.55) (0.02) West Bengal 52.72 1.143* .30 .190 0.84** .37 (9.073) (2.05) (5.20) (2.41) *Indicates that the coefficient on Time is significant with the sign indicated at the 10 percent level for a two-tail test. **Indicates significance at the 5 percent level for a two-tail test.  20 30. The main conclusions to be drawn from these results is that the incidence of poverty in rural India does not show any sustained trend over the past two decades. What we observe in most of rural India is a pattern of fluctuation with an increase in the incidence of poverty in one sub-period followed by a decrease in another. This pattern obviously calls for some explanation in terms of the economic factors affecting the rural economy and this is attempted in Section V. But first it is worth exploring the implica- tions of using national accounts data to supplement the NSS data. IV. ALTERNATIVE POVERTY ESTIMATES USING NATIONAL ACCOUNTS DATA 31. In the preceding section, we have relied solely upon the NSS con- sumption surveys as the principal source of information on the level and distribution of consumption in rural areas. There is, however, an alterna- tive approach used by Minhas (1970), which relies upon the national accounts series of private consumption expenditure to estimate the level of consumption per person in the rural areas and then uses the distribution observed in the NSS surveys to estimate the incidence of poverty. Since there are substantial differences between the National Accounts estimates of the growth of per capita consumption and estimates derived from the NSS, this approach produces a very different assessment of trends in poverty in rural India. In this section, we will present the results obtained from this approach using the recently published official National Accounts series and then assess them against the results obtained from the NSS. 1/ 32. Minhas' procedure consists of estimating the ratio of rural consump- tion per person to consumption per person in the whole economy and applying this ratio to the time series of consumption per person in the economy obtained from the national accounts. Table 5 presents national accounts based time series of rural consumption levels following this approach, for the years 1960/61 onwards. 2/ The discrepancy between the national accounts and the NSS estimates of private consumption expenditure in the economy as a whole can be seen by comparing columns 4 and 5. The two estimates are very close in 1960/61 but diverge sharply thereafter, with the national accounts series showing a much faster growth in consumption per capita. This discrepancy at the national level is also reflected in our national accounts based estimate of rural per capita consumption in current prices (column 7), which shows faster growth than the time series of direct estimates from the NSS (column 2). 33. These national accounts based estimates of rural consumption per person can be used to estimate the percentage of the population in poverty as well as the Sen Index on two different assumptions. We may take our current price estimates of rural consumption expenditure per person (column 7) as the estimate of mean expenditure and the estimated Lorenz curve based on the NSS distribution, and then estimate poverty measures using the line in current 1/ The national accounts in current and constant prices have been recently revised and are reported in Central Statistical Office (1975). 2/ The official national accounts data on consumption are available only from 1960/61 onwards.  Table 5 ALTERNATIVE ESTIMATES OF RURAL CONSUMPTION PER PERSON National Accounts Esti- National Accounts Based mates of Consumption Per Estimates of Rural Con- Direct Estimates of Con- Person for 30 Days 1r sumption Per Pe son for sumption Per Person for 30 Days Economy as a Whole 30 Days_' (Rs. in Current Prices) Rs. in Rs. in Rs. in Rs. in Ratio of Urban to Whole 1/ Current 1960-61 Current 1960-61 Total Population Rural Urban Economy- Prices Prices Prices Prices (1) (2) (3) (4) (5) (6) (7) (8) 1960-61 .1798 21.47 29.52 22.92 22.67 22.67 21.23 21.23 1961-62 .1812 21.73 30.86 23.38 23.17 22.22 21.53 20.65 1963-64 .1843 22.37 32.96 24.32 26.10 22.41 24.00 20.62 1964-65 .1860 26.44 36.03 28.22 30.39 23.91 28.48 22.40 1965-66 .1879 26.55 35.45 28.22 31.40 22.82 29.54 21.47 1966-67 .1899 30.90 41.54 32.92 36.14 23.15 33.92 21.72 1967-68 .1920 33.40 44.82 35.59 42.65 24.16 40.02 22.68 1968-69 .1942 33.28 46.04 35.76 41.64 24.53 38.75 22.83 1970-71 .1991 35.31 52.85 38.80 46.41 25.81 42.24 23.49 1/ Figures in these columns are obtained as a weighted average of Rural and Urban consumption per person using the Rural and Urban proportions of total population as weights. 2/ Calculated from estimates of annual per capita private final consumption expenditure as reported in "National Accounts Statistics 1960-61 - 1974-75", Central Statistical Organisation, Government of India, October 1976. 3/ These estimates are obtained by using the NSS ratio of rural to total consumption (column (2) column (4)), and applying this ratio to the National Accounts Estimates of consumption per person in current and constant prices.  22 prices based on the CPIAL. Alternatively, we can use the constant price estimates of consumption (column 8) and-the same Lorenz curve and estimate the poverty measures taking the poverty line fixed at Rs 15 in 1960/61. The second procedure is equivalent to assuming that the price index relevant for the poor is the same as the deflator for private consumption in the national accounts. 2/ 34. Table 6 presents estimates of our two poverty measures on each assumption. As we would expect, the estimates of poverty are very close to the NSS based estimates in Table 2 for the year 1960/61, but these estimates are much lower than the NSS estimates for later years. This ref'lects the faster growth in mean consumption levels implied by the national accounts. It is interesting to note that even if we adopt the national accounts based time series of consumption per capita, the choice of a price index is crucial. Thus, the estimates in Table 6 show a steady decline in the incidence of poverty, whether we look at the percentage poor or at the Sen Index as our measure of the incidence of poverty. As noted above, these estimates are based on the assumption that the prices for the rural poor rose at the same rate as the private consumption deflator. By contrast, the estimates based on the current price consumption series using a poverty line based on the CPIAL shows a cyclical pattern with a distinct worsening in the mid-sixties and an improvement thereafter. 35. Choosing between estimates based on different data sources is always a difficult task and one can do little more than state one's prefer- ences and the reasons behind them. In our view there are good reasons for favoring the NSS based estimates of poverty over those obtained by combining the NSS with National Accounts. The NSS estimates are direct estimates of rural consumption levels based on scientifically designed and well implemented surveys. The National Accounts estimates, on the other hand, are based on the commodity flow method and incorporate a very large element of discretion in allocating estimated supplies of commodities to alternative user categories of demand. Furthermore, in order to obtain a series on rural consumption from the National Accounts it is necessary to use the ratio of rural consumption per person to consumption per person in the whole economy. This procedure raises obvious questions about the reliability of the NSS ratio given our unwillingness to use the NSS estimate itself. 1/ Given the NSS distribution, a higher mean consumption necessarily implies a lower estimate of the percentage of the population in poverty. This can be seen from Figure 1, where it is evident that since a higher mean income implies a lower ratio of the poverty line to the mean income, the point x must move lower down the Lorenz curve for the slope of the curve at x to be equal to this ratio. 2/ Minhas used the GDP deflator because a separate consumption deflator was not available. However, the revised national accounts series provide a separate deflator for private consumption and this is clearly to be preferred.  23 Table 6 NATIONAL ACCOUNTS BASED ESTIMATES OF POVERTY MEASURES Based on Current Price Series of Con- Based on Constant Price Series of Con- sumption Per Person Using Poverty sumption Per Person (in 1960-61 prices) Lines Based on CPIAL Using Constant Poverty Line of Rs. 15 Percentage of Rural Sen's Poverty Percentage of Rural Sen's Poverty Population in Poverty Index PoDulation in Poverty Index 1960-61 39.7 .146 39.7 .146 1961-62 40.1 .143 41.0 .148 1963-64 38.6 .131 37.5 .126 1964-65 40.7 .143 31.9 .103 1965-66 45.0 .166 35.1 .119 1966-67 49.6 .198 34.8 .121 1967-68 42.5 .156 30.7 .100 1968-69 38.9 .137 32.2 .106 1970-71 33.5 .106 28.5 .085  24 36. Quite apart from the choice between the NSS estimates and the hybrid NSS cum National Accounts estimates of consumption levels, there is the sep- arate question of whether the poverty line in current prices should be defined on the basis of the consumption deflator in the national accounts or the Con- sumer Price Index of Agricultural Laborers (CPIAL). In view of the accumulated evidence suggesting that the price of food items (especially coarse grains) rose faster than other prices, there is clearly a strong argument in favor of the CPIAL. 1/ As shown above, the use of the CPIAL based poverty line sub- stantially alters the picture of trends in poverty compared to the results obtained by using the consumption deflator, even if we use the NSS cum National Accounts series on consumption. There is no longer a trend decline in the incidence of poverty. Instead, there is a pattern of fluctuation very similar to that reported in Table 2. 37. We conclude that the results reported in Minhas (1970) showing a decline in the percentage of rural population in poverty in the sixties probably present an over-optimistic picture. What is more likely is that the incidence of poverty fluctuated over the period. Certainly over the sixties it is very likely that the incidence of poverty actually increased. V. AGRICULTURAL PERFORMANCE AND RURAL POVERTY 38. In this section we attempt to explain the observed fluctuations in the incidence of poverty documented in the preceding sections. Ideally, the changes in poverty in different subperiods should be explained in terms of some explicit model of the determinants of rural poverty and the behavior of these determinants over the subperiod. In this context, it is obviously relevant to consider the rate of agricultural growth, the factors determining its distribution across farm sizes, its impact on tenancy conditions, its effect on the demand for labor, etc. A complete exploration of the impact of these factors on the economic condition of different socio-economic groups in the rural economy is obviously beyond the scope of this paper. Instead, we shall confine ourselves to examining the relationship between the incidence of poverty and some aggregate indices of agricultural performance over the period 1956/57 to 1973/74. Once again the analysis is presented separately for India as a whole and for the individual states. (A) The All-India Evidence 39. We begin by postulating that an important determinant of the extent of rural poverty is the level of agricultural production relative to the size of the rural population. Agriculture is not the only source of income in rural areas but it is the dominant source, and besides, the scale of non- agricultural income generating activity in rural areas almost certainly de- pends upon the level of agricultural production. If there is any "trickle down" mechanism at work in the rural economy, we should expect increases in agricultural production per head to reduce the incidence of absolute poverty. Does the available evidence support this view? 1/ See Bardhan (1971). For an opposite view see also Minhas (1971).  25 40. The first point to note about the Indian experience of the past two decades is that the growth of agricultural output has only just about kept pace with the growth of the rural population. The resulting stagnation in agriculture can be seen from Table 7, which presents time series of three different indices of agricultural output per head of the rural population with output being measured alternatively by the index of foodgrains production, the index of agricultural production (all crops) and the Net Domestic Product (net value added) in agriculture in constant.1960f6.1 -pr-ice-s. None o-