The Impact of the Non-Farm Sector on Earnings and Gender Disparities in India:

The Impact of the Non-Farm Sector on Earnings and Gender Disparities in India: 1983-99 Mukesh Eswaran #, Ashok Kotwal #, Bharat Ramaswami *, & Wilima Wadhwa $ April 19, 2005 Preliminary draft prepared for World Bank workshop on Equity and Development, December 7-8, 2004. We are grateful to Shastri Indo-Canadian Institute for financial support through SHARP. #The University of British Columbia, Vancouver * Indian Statistical Institute, Delhi $ India Development Foundation, Gurgaon

The Impact of the Non-Farm Sector on Poverty Reduction and Gender Disparities in India: 1983-99 1. Introduction This paper is a part of a larger project that aims to assess the impact of the liberalization measures undertaken in the decades of the Eighties and Nineties on earnings and gender disparity across India. In the first part of this paper, we investigate whether and how the growth in the industrial and service sectors may have been responsible for the observed increase in the earnings of wage earners during the time span of 1983 through 1999. Our attempt here will be to take a careful look at NSS Employment surveys to see if we can speculate on what might be the processes responsible for the increase in earnings. Is it mostly the agricultural productivity increase or is it the growth in non-agricultural sectors (industry and services) through the creation of employment opportunities for the poor outside agriculture that may be causing the increase in labour earnings? We also examine the role of education in the process of poverty removal and the differential impact of the growth process on women in the labor force. 2. Literature Studies of poverty decline have emphasized the composition of growth as a determinant of the extent of trickle-down. Ravallion and Datt (1996) found rural economic growth to have significant impact in reducing urban and rural poverty while urban growth has little impact on rural poverty. In the rural sector, higher farm yields is the key variable that reduces poverty and increases wage earnings (Datt and Ravallion, 1998). On the 1

other hand, the impact of non-farm economic growth varies across states (Datt and Ravallion, 2002). In particular, the authors find that non-agricultural economic growth was less effective in reducing poverty in states with poor initial conditions in terms of rural development and human resources. Thus, low farm productivity, low rural living standards relative to urban areas and poor basic education and high infant mortality and greater landlessness all inhibited the prospects of the poor participating in the growth of the non-farm sector. These findings, therefore, suggest a synergy between rural development and the role of the non-farm sector in poverty reduction. In a recent study, Foster and Rosenzweig (2003) question the complementarity of agriculture and the non-farm sector in the rural growth process. They argue that it is only the non-traded segment of the rural non-farm sector that is positively affected by agricultural growth. The traded non-farm sector (consisting of rural factories) competes with the farm sector for labour and is therefore attracted to regions with low agricultural productivity. Their empirical results show that much of the growth in the rural nonfarm sector in India was due to the rural factory sector and not due to the non-traded sector. Furthermore, the growth in rural factory sector was twice as important as farm yields in accounting for the increase in rural wages over the period 1982-99. This paper examines the all-india employment and earnings data for the period 1983-99. Our goal is to see whether the evidence is supportive of the independent and powerful role of the non-farm sector that emerges in the Foster and Rosenweig analysis. Our analysis is based on the employment surveys conducted by the National Sample Survey Organization (NSSO). While these data are publicly available, Foster and Rosenzweig use a data set that is not available to us. We should note that Foster and 2

Rosenzweig use the agricultural wage as their dependent variable. It is indeed a good indicator of poverty (the correlation between povery and agricultural wages is displayed graphically in Deaton and Dreze (2003)) and easier to deal with. In this paper, we too will therefore focus on agricultural wage or earnings (daily and weekly) rather than any poverty index. Also, since the agricultural wage in India is the reservation wage for workers in other sectors, it is also an important determinant of wages in industry and service sectors. 3. Conceptual Framework What are the different channels through which there could be an increase in the agricultural labor incomes? A natural framework for thinking about these issues is a two-sector framework consisting of agriculture and industry 1. Agriculture uses land and labor and other inputs. Increases in the marginal productivity of labor either through technological change (e.g., better seeds) or through investment (e.g., irrigation) in agriculture will, of course, directly increase labor incomes or earnings. If an increase in industrial productivity lowers the cost of an agricultural input such as fertilizers or pesticides thereby causing an increase in agricultural productivity, then that would be one avenue through which industrial productivity increase can raise incomes of agricultural laborers. Fertilizer prices in India, have, however been controlled by the government and there is no room therefore to believe that this has 1 See Eswaran and Kotwal (1993) for a precise model on what our framework is based. 3

been the avenue through which growth in industrial labor productivity might have had an impact on the incomes of the rural poor. The input-supplying sector that has been fundamentally altered by economic reforms is the seed industry. Because of the removal of industrial licensing and the curbs on direct foreign investment, this sector has seen increased investment spending and entry by large foreign majors. However, the presence of the private seed industry in India s principal food crops, rice and wheat, is insignificant. Thus, it is unlikely that agricultural labor incomes could have increased via this channel. It is theoretically possible that productivity growth in non-agricultural sector can lower the relative prices of industrial goods and services in the consumption basket of an agricultural laborer and improve his real wage. However, an examination of the weights given to different items in an agricultural laborer s basket (the CPI for Agricultural Laborers, Labor Bureau) brings out the fact that about 69% of the expenditure in 1983 is on Food. Only 16% of the consumption expenditure can be classified as those products of either industry or service sector whose prices are to some extent market driven. These include Intoxicants (3.8%), Clothing and Footwear (6.7%), and Miscellaneous (5.7%) (i.e., Transportation, Toiletries and Others) where the figures in the parentheses indicate the expenditure percentages in 1983. During the period 1983 through 1999, the consumer price indices for these commodity groups have increased by 2.16 times for Fuel and Light, 4.05 times for Clothes and Footwear, 6.38 times for Miscellaneous while it has increased by 3.3 times for Food. Thus there is no evidence that the relative prices of the non-agricultural items that an agricultural laborer would consume have fallen 2. 2 This despite the fact that the terms of trade turned in favor of agriculture during the decades of the 80 s and 90 s. 4

An important avenue through which the growth in non-agricultural productivity could affect agricultural wages is by lowering the labor-to-land ratio in agriculture. Because of diminishing returns, agricultural wages (for a given level of productivity) are inversely related to the labor-to-land ratio that, in turn, depends on the capacity of non-agricultural sectors to draw labor from agriculture. Thus, when the expansion of non-agricultural sector results in a movement of labor away from agriculture, it not only confers benefits on the labor that moved (through perhaps higher wages in nonagriculture) but to all those still left in agriculture. This is the main conduit through which non-agricultural growth can have an impact on rural poverty in a country like India. The Asian Miracle in countries like Singapore, Hongong, South Korea of the 70 s and 80 s was based on precisely such a process 3. It is fruitful therefore to examine how much has the labor-to-land ratio changed in India from 1983 to 1999. The labor-to-land ratio is, however, an endogenous variable that is driven by the productivity increase in either agricultural or non-agricultural sector. For example, in a closed economy, an increase in agricultural productivity will make food cheaper and if the income elasticities of non-agricultural goods are higher than for food (Engel s Law) then the demand for non-agricultural goods will increase. In the general equilibrium, labor will move to non-agriculture to satisfy the extra demand for non-agricultural goods. On the other hand, in a small open economy, an increase in industrial productivity can also result in an expansion of the industrial sector and hence a movement of labor from agriculture to industry. Thus since a movement of labor from agriculture to non-agriculture can be caused by an increase in either agricultural 3 Of course, in several other countries like Taiwan and Indonesia the increases in agricultural productivity 5

productivity or industrial productivity, attributing the entire decrease in the labor-toland ratio solely to industrial growth overestimates the contribution of non-agriculture towards raising the agricultural wages. Separating out the relative contributions of sectoral productivity change on agricultural earnings is empirically difficult because of the absence of such productivity measures at a suitably disaggregated level. In later sections, our empirical strategy would be to bound the contribution of non-agriculture by in fact ascribing the entire change in labor-to-land ratio to non-agriculture. 3. Data Our data sources are the employment surveys of the NSSO. In this paper, we consider the survey in 1983 (calendar year) and in 1999/00 (agricultural year i.e., July 1999 to June 2000). The survey adopts a two stage sampling design first the primary sampling units are randomly picked (villages in rural areas and blocks in urban sector) and then households are randomly chosen. At each stage, the survey is stratified as well. At the first stage, stratification is according to population. At the second stage, all households are stratified into 2 strata: affluent households and the rest. Table 1 provides information about the size of the sample in each of these years. The survey period is divided into 4 quarters and the sample design allots equal number of primary sampling units to each quarter. Thus, for instance, about 30,000 households were surveyed in each quarter of the 1999/00 survey. The survey data do not report the day or week when the household is surveyed although the instructions for fieldwork state preceded the industrial expansion and also played an important role in increasing rural wages. 6

that within a quarter the fieldwork is spread uniformly over the different weeks. Note that the uniform allocation of household units across sub-rounds applies at the level of the state as well. Thus, in comparing outcomes at the state-level across NSS rounds, we can be sure that we do not have to adjust for seasonal factors. For a given reference period, individuals are classified as being in the work force, unemployed or being out of the labor force. When the reference period is a year, the `usual status of an individual is determined on the major time criterion. For an individual who is employed on the usual status, their principal activity in terms of industry of employment is also determined on the basis of major time criterion. The survey also records their `subsidiary economic activity in the remainder time. For the reference period of a week, the survey elicits an individual s time disposition during each day of the week. For each day, individuals are classified (their `daily status) as being in the work force, unemployed or being out of the labor force with a weight of either 1.0 or 0.5. A weight of 1.0 corresponds to a full day and a weight of 0.5 corresponds to a half-day. Naturally, an individual can at most be assigned two activities with equal weight. The survey uses a priority & major time criterion to assign the activity status to each half-day. This is explained in Table 2. Summing the weights across days, we obtain for each individual in the survey, the weekly break up of days in each of the 3 activity states. For assigning the industrial classification code, a person who is considered to be employed for the day would be assigned at most two economic activities (with weights 0.5 apiece) decided on the major time criterion. A person who is employed for half-day only would be assigned one economic activity again on the major time criterion. Once 7

again by summing the weights across days, we obtain for each individual in the survey, the weekly breakup of the days of employment into different economic activities. For the reference period of a week and for each economic activity reported by an individual, the employment survey also reports the weekly earnings. A measure of daily earnings in the activity can be obtained by dividing the weekly earnings by the number of days worked in that particular activity. In this paper, we work with the data on the weekly disposition of time and the `daily status classification of economic activities into various industrial classifications. The data on the weekly disposition of time takes into account multiple economic activities that are characteristic of poor households. Furthermore, as households are surveyed throughout the year (in equal numbers), the aggregates derived from weekly data are representative of annual aggregates. 4 Sectoral Shifts in Employment: Agriculture s Share Tables 3 and 4 show the changes over the 16-year period (1983 to 1999) in the employment structures for males and females respectively. The table is based on the one digit daily status classification of economic activities. It presents the shares of 4 sectors: agriculture; manufacturing; the aggregate of construction & trade & hotels, and transport & communications (CTT); and lastly the aggregate of government services, health, education and various personal services (G&P). Employment shares of mining and of real estate and finance are not presented which is why the total of shares adds to a number slightly below 1. 8

The all India employment share of agriculture has declined by 7 percentage points for males and by 2 percentage points for females. Non-agricultural jobs opening up in the 80 s and 90 s seem to favor men over women. At the level of states, the experience is diverse. Consistent with the aggregate outcome, there is a large group of states that have experienced no structural change. Kerala, TN are the states where the share of farm sector has declined sizably for both genders. Assam, Haryana, Kerala, Punjab and Rajasthan records substantial (over 12 percentage points) decline for men. Andhra Pradesh, Gujarat, Maharashtra, Tamil Nadu, and UP have between 5 and 8 percentage points decline in the share of agriculture. The rest of the states have less than 3 percentage points decline in their agriculture share of employment. Kerala is the only state where there has been a decline of more than 5% in the agriculture share of employment for women. Incredibly, Punjab shows a hefty increase in the agriculture share. What are the non-agricultural sectors that have expanded in relative terms. Consider first the figures for males. Of the states that show a substantial decline in agriculture s share, it is only in Haryana that the share of manufacturing increases substantially from 10 to 17 percentage points. The share of manufacturing declines in Andhra Pradesh, Assam, Bihar, Gujarat, Kerala, Madhya Pradesh, Orissa, Rajasthan, and West Bengal, while it increases by 3 percentage points in Punjab, by 1-2 percentage points Tamil Nadu and Uttar Pradesh. In all of these states, the sector that expands the most in relative terms is construction, trade and transport. The other feature of Tables 3 and 4 that is worth noting is the decline of the government and personal services sector. Overall it has declined by more than 2 9

percentage points and it declines in every state except Assam. Even in states where agriculture s share has declined little, there is a reallocation of employment from the government and personal services sector to construction, trade and transportation. This happens in states as disparate as Bihar, Gujarat, Karnataka, Maharashtra and Orissa. This suggests that there may be different sources of income growth in different states (e.g., remittances for Kerala, government programs for Assam, agricultural growth for Rajasthan, manufacturing growth for Tamil Nadu, etc) and the increases in income may be spilling into an increase in demand for services such as construction, trade and transport. For women, the remarkable feature of the employment structure is its constancy over time in the aggregate and for most of the states. Unlike the case with men, the share of the government and personal services sector does not fall. In some states like Karnataka, Orissa, the share of female employment in agricultutre increased, and quite drastically in Punjab. Of the states that experienced a fall in agriculture s share, Tamil Nadu, Uttar Pradesh and West Bengal are the only states where there is a rise in the share of manufacturing in the female workforce. Note that such a development was not seen for males in West Bengal suggesting that the rise in the female proportion does not reflect large-scale industrialization. Kerala s pattern mirrors that of the men as construction, trade and transport absorb the decline in agriculture s share. 5. Labor-to-Land Ratio Since the population and hence the labor force was increasing in India during the 80s and the 90s, a decrease in the agriculture s share of employment, of course, does not 10

mean that the labor-to-land ratio declined too. It is instructive to examine what happens to the labor-to-land ratios across the country. Figure 2 shows that in all but 4 states (Haryana, Kerala, Punjab and Rajasthan) this ratio has increased from 1983 to 1999. 4 Note that even in these 4 states, the fall in the labor-to-land ratio was very small. In all other states, the agricultural wages increased despite the increase in this ratio, implying that the agricultural productivity growth more than compensated for the downward pressure on wages caused by the agricultural sector absorbing more labor. Clearly, if the agricultural productivity had not increased, the agricultural wage would have in fact decreased (in all but these 4 states). The positive contribution of agricultural productivity growth is self-evident. However, if the non-agricultural sector had not created any jobs during the time interval (1983-1999), all the additions to the labor force would have had to be absorbed by the agricultural sector and the wages would have been even lower. This is, of course, a completely fictitious scenario that is useful only as a thought experiment that may allow us to ask what is the maximum contribution that the non-farm sector could have made to the labor incomes in the agricultural sector. Table 5 allows us to examine such a counterfactual. Table 5 presents the person days of employment created by the non-agricultural sector during 1983-99. This is expressed as a percentage of the total agricultural employment in 1999. Note that there is considerable variation in the non-agricultural employment created between 1983 and 1999 across states. Putting aside the outlier 4 The numbers on land are taken from official statistics on gross cropped area in the early 1980s and in the late 1990s. The source for the former is Bhalla and Tyagi (1989) and for the latter is the website www.indiastat. com. Labor totals are derived from the NSS Employment Surveys (38 th and 55 th rounds). 11

Kerala (45% of total agricultural employment created in the non-agricultural sector from 1983 to 1999), the range of variation is from 10% in Orissa to 30% in Punjab. If we focus only on the additional days of non-agricultural employment created for workers with low levels of education (no more than secondary) the numbers are much smaller: they range from 4% for MP to 23% for Assam. It is noteworthy that the states that are well-known for having attracted relatively greater investment in the modern industrial sector (i.e., Gujarat, Haryana, Karnataka, Maharashtra and Tamil Nadu) have created considerable employment in their non-farm sectors but no more than around 17% of the addition is for workers with low level of education. It is these workers rather than educated workers who would have worked in agriculture if they had found no nonfarm employment. To take our thought experiment further, we need to be able to answer: how much lower would the observed change in labor earnings in agriculture have been had none of the additions to the labor force been employed in non-farm activities? To answer this question, we assume that the agricultural output, Q, is given by a Cobb-Douglas production function: Q = AL α H β, with α + β = 1, where L and H, respectively, denote the amount of labor and land employed in agriculture, and A denotes the total factor productivity. Let the period subscripts 1 and 2 denote variables in the years 1983 and 1999, respectively. Denote by Li, H i, Ai, wi ( i = 1,2), the amount of agricultural labor used, land used, the total factor productivity, and the agricultural wage, respectively, in period 12

i. Taking the agricultural output as the numeraire, the wage rate (given by the marginal product of labor) may be written w i = β Aα ( L / H ), i = 1,2. i i i The exponent of the labor-to-land ratio on the right hand side of the above expression is easily shown to be the elasticity of the wage rate with respect to labor-to-land ratio. The wage rates (approximated by earnings in this exercise) are observed in each of the two periods. Suppose l denotes the increase in non-farm employment over the 16 years. Assume that if the non-farm sector had not created this employment, the employment in agriculture in 1999 would have been higher by this amount. Let H w 2 denote the agricultural wage that would have prevailed in 1999 in this hypothetical scenario. Assuming the same amount of land would have been employed in this fictitious setting as was actually employed in 1999, we may write β w H 2 = A2α[( L2 + l) / H 2 ], which may be rewritten β β β w H 2 = A2α ( L2 / H 2 ) (1 + l / L2 ) = w2 (1 + l / L2 ). The observed increase in wages over the 16 years is w2 w1. The increase in wages that would have been observed in the hypothetical scenario is given by w H 2 w1 = w2 w β ( 1+ l / L2 ) 1. The total increase in wages can be decomposed into that which would have obtained if the non-farm sector had not absorbed any additional employment plus the increase that came about because of additional non-farm employment: 13

w 2 w1 w H = ( w2 w1 ) + ( w2 1). H Using the expressions for and from above, we obtain the proportional increase in w 2 the agricultural wage attributable to non-farm employment as: w 2 H ( w2 w2 ) / w1 = ( w2 / w1 )[1 (1 + l / L2 ) This expression is couched entirely in terms of observable quantities and can be readily computed. It is worth noting that the above expression increases with agricultural productivity growth during the period since the ratio of wages across the two periods appears as a multiplicative term. The contribution of non-farm sector is boosted by an increase in agricultural productivity. β ]. To estimate the right hand side, we need to have some idea of what the value of the elasticity parameter β ( = 1 α) is. Table 6 gives the labor share in the cultivation costs of some major crops for different states. The higher the labor share, the lower is the elasticity. In order to give the maximum benefit of the doubt to non-farm sectors in the assessment of their contribution, we pick a number that is less than the lower bound in labor shares, i.e., we set α = 0.15, so that β = 0.85. We then evaluate the maximum possible increase in wages of labor across all educational classes that could be attributed to non-farm sectoral growth using an elasticity of 0.85. This is done in Table 7. The first column here is the observed change in agricultural earnings in percentage terms. The second column is the maximum projected change in agricultural earnings that could be possibly attributed to the growth of the non-farm sector. It is computed by using the methodology shown above. The third column derives the contribution of non-farm sector to overall increase in agricultural earnings as the ratio of the projected earnings to 14

the observed earnings. In the next 3 columns, we repeat the exercise for workers with low level of education (i.e., no more than primary). Non-farm sector has clearly played a significant role in Haryana with an upper bound of the %contribution of 93% in the increase of agricultural labour earnings respectively. In Kerala, Gujarat and Punjab too the non-farm sector has played a significant role in the agricultural earnings growth. However, in all other states the non-farm sector gives us no more than 50% of the earnings increase as its maximum possible contribution. Under a vastly more realistic assumption that only labor with at most secondary education would accept working in agriculture, once again Haryana and Kerala show a significant contribution of non-farm sector 56% and 59% respectively. In the rest of the states the maximum contribution is 42% in Punjab and the lowest is close to 10% in MP for those with no more than secondary education. It is difficult therefore to believe that in most states the contribution of the non-farm sector to the growth in incomes of agricultural laborers has been significant (>35%). 6. Employment Shifts: Cohort Effects First, we examine the sectoral patterns of employment (at the 1-digit level) disaggregating the population into cohorts of 8 year age intervals. Tables 8(a) and 8(b) give the absolute amount of employment in person days (million)of employment and the distribution of employment shares in the total employment for males in 1999 and in 1983 respectively. Tables 9(a) and 9(b) give the same information for females. Note that the 1983 cohort 18-26 becomes the 1999 cohort 34-42 sixteen years later. Similarly, the 15

1983 cohort 26-34 becomes the 1999 cohort 42-50 and so on. It is evident that it is only the youngest 1983 cohort of males (18-26) that shows a significant decline in the agriculture s share in employment. In other words, it is this group that is finding employment in industry and services. Neither the older cohorts nor females show a significant decrease in the agriculture s share of employment. After the age 26-34, employment patterns tend to get frozen. The older cohorts do not benefit a great deal from non-agricultural employment. The cohort 58-66 in fact shows an increase in agriculture s share of employment as the days of employment in agriculture remains unchanged while that in non-agriculture drops dramatically. As for women, we don t observe the shifts out of agriculture even in the younger cohorts. The reason that the employment days for the youngest cohort increase so much over the sixteen years is that those outside of the labor force in 1983 come into the labor force in 1999. The total employment numbers decline for older cohorts but they drop even more significantly in non-agriculture. It seems clear that the job market in industry and services has favored younger people and that too men. 7. Education and the Role of the Nonfarm Sector The previous section suggested that the shift out of agriculture is associated with education. To make this connection explicit, this section considers the role of the nonfarm sector in the earnings of workers differentiated by their education levels. A well known feature of earnings data is that even after controlling for education and age, earnings differ between industries. In India, earnings in agriculture are typically the lowest. Suppose W 0 is the expected earnings of an illiterate in 1983. Then 16

(1) W0 = p0iw n i= 1 0i where is the average earnings in sector i, is the probability of obtaining w 0 i p 0i employment in sector i and n is the number of sectors. Similarly, if W 1 denotes the expected earnings of an illiterate in 1999, then (2) W1 = p1iw1 i n i= 1 Notice that expected earnings in 1999 could be different from that in 1983 either because of an increase in sectoral earnings or because the sectoral probabilities of employment change or both. If the agricultural sector is indexed by 1, then the contribution of this sector to the total income of the illiterates in each year is given by (3) ρ 0 = p01w01 /W0 and ρ 1 = p11w11 /W1 To obtain the estimates of (1), (2) and (3), we do the following at the all-india level: (a) We compare the earnings of workers without schooling of cohort aged 34-42 in 1983 with the earnings of workers without schooling of also cohort 34-42 in 1999. We do this separately for males and females. (b) We also compute the sectoral distribution of workers (based on the daily status classification) in the relevant sub-population (for e.g., illiterate female workers of cohort aged 34-42 in 1999). We use the sectoral proportions as probabilities of employment. (b) The results are displayed in Tables 10 & 11. 17

For illiterate males, agricultural activity accounts for 68% of working days in 1999 and 68% in 1983 in contrast to the ratio of 60% for the population as a whole. Notice that the sectoral distribution is virtually unchanged between the two years. The increase in expected earnings is therefore entirely due to higher sectoral earnings and none at all due to sectoral shifts of employment. Among the non-farm sectors trade & hotels command the least industry premium. Mining commands very high premium but employs very few people. Manufacturing and Transportation command over 100% premium. Construction is somewhere between Trade and Transportation. Surprisingly, the contribution of agriculture to the total earnings of this group has increased only slightly, from 56% to 57%, in keeping with the fact that the share of employment in agriculture has fallen because of the slower rate of increase in earnings in the nonfarm sector. For illiterate females, the contribution of agriculture to their total income is much higher of the order of 72 and above. Between 1983 and 1999, their dependence on agriculture for employment has increased from 63 % to 80%. Even though non-agricultural earnings have increased at a slightly faster pace, it has been more than compensated by an increase in the employment share of agriculture from 76% to 80%. We, therefore, see that the nonfarm sector has played a limited role in accounting for the higher earnings of male illiterates and none at all for female illiterates. How does the impact vary with education level? To answer this, we repeat the exercise in the earlier section for individuals who have completed middle or secondary school. The 18

results are displayed in Tables 12-13. Note that here too the earnings figures excludes the self-employed. Notice that the contribution of agriculture drops dramatically for middle school completed males and females. Note that it is lower than the agriculture s share of employment because of the much higher earnings in other sectors. There is something else noteworthy here: 93% of males in this group were employed in sectors other than agriculture in 1983 whereas only 83% of them were so employed in 1999. This is surprising since the non-farm sectors are expected to have created employment for this group during the Eighties and Nineties. Indeed, manufacturing, construction, trade, transportation have all increased their share of employment over the time period. It is the government service that has dropped its employment share from 49% in 1983 to 24% in 1999. This is what is primarily responsible for the reduction of the contribution of nonfarm sectors in the total earnings of this group. Within the nonfarm economy, 4 sectors account for most of the expected earnings. These are manufacturing, communications & transport, real estate and finance and the sector consisting of government, social and personal services. 19

8. Educational Premia In the last section we saw that there exists an industry premium in non-farm activities over what a worker with certain age and education characteristics can get in agriculture. It pays to get non-farm jobs and the probability of getting these jobs rises with education. In trying to assess the contribution of the growth in non-farm sectors toward poverty removal, we can ask the following important question: would the contribution have been much greater if much greater part of the population was educated? In other words, where is the bottleneck in the rate at which educated work force is being generated or in the rate at which employment opportunities are being created? We can get some idea by looking at what is happening to the educational premia over time. To capture this educational premium we ran the following regression: lnwij 0 + Β1' E ij + Β 2 ' Cij + B 3 = β ' N + δ + ε ij j ij where i indexes the individual and j indexes the state, W is earnings, E is a vector of dummy variables indicating the individual s education level, C is a vector of dummy variables for the individual s cohort, N is a vector of interaction variables between the education and cohort dummies and δ is a fixed effect specific to the state. Since there are 6 educational classes in the 55 th round and only 5 in the 38 th round we have collapsed the educational classes into 4 classes that would be compatible across the two rounds: (1) Illiterates, (2) Primary, (3) Middle school, and (4) Graduates (High school graduates and also University graduates). The coefficients on educational dummies allow us to determine the educational dummies for each cohort. An illiterate worker belonging to the cohort 3 (i.e., age group 34-42) had an all India average weekly earnings of Rs. 133 in 1983 while for a worker with primary education the figure was 20

Rs. 162. Thus, the educational premium for primary education was Rs. 29.5 Similarly, the educational premia for Middle school and Graduates over illiteracy were Rs. 101and Rs. 240 respectively. The results for the 55 th round show that these premia have increased to Rs. 53, Rs 143 and Rs. 459 respectively. For the next older cohort, the increase in premia is even greater. What this indicates to us is that if more Middle school and High school graduates were available in 1999 they would have found employment in industry and services. The main reason why the non-farm sector has not been able to contribute more to the poverty removal is that most of the employment it creates is for educated workers rather than for the illiterates and primary school graduates. 9. Women and the Non-Farm Sector From the previous sections, we know that the non-farm sector jobs have largely favoured men and that too in the younger cohort. Employment proportions for women have remained stagnant and that picture holds for younger cohorts as well. This could possibly be due to educational disparities between the genders as we find entry into nonfarm sector to be highly correlated with education. The absence of a structural shift in women s employment patterns raises the question whether women have received even the limited gains (from the growth of nonfarm sector) that have accrued to men. This could happen if the greater competition for male labor results in a higher demand for female labor in agriculture. Thus as male labor leaves agriculture, female labor takes its place and gains from declining labor-land ratio. 21

The three principal activities in agriculture on which most male labor is expended (67% of total male labor in 1983 and 66% of total male labor days in 1999) are other cultivation activities, harvesting and ploughing. If female labor substitutes for male labor then we should observe that the ratio of male to female labor declines in agriculture especially in the principal `male agricultural operations. Table 14 presents the ratio of total male labor days to female labor days in various agricultural operations for 1983 and 1999. Notice that for all operations except harvesting, female labor input has increased relative to male labor. Overall the ratio drops from 2.25 to 2.02. Thus, the evidence supports the hypothesis that agriculture uses more female labor to fill the jobs vacated by male labor. It would then seem that despite the absence of shifts in the employment structure, women have also obtained the gains from a growing non-farm sector. Indeed weekly earnings for women in the agricultural operations listed in Table 14 have grown at about the same rate (49%) as that of men in similar activities (53%). 10. Conclusions Our results at this point can only be suggestive. Our examination of the NSS Employment Surveys indicates that the contribution of the growth in labor productivity in the non-farm sectors toward an increase in agricultural labour earnings during the period 1983 through 1999 in most states is likely to have been quite limited. Haryana, Kerala, Gujarat and Punjab are the notable 22

exceptions. It is likely that it is the growth of agricultural productivity that has played a major role in raising the agricultural wages and thus in reducing poverty. The non-farm sector has created jobs for literate people and the younger cohorts who are able to raise their educational status and move out of agriculture. Given that the educational premia have increased over time, it seems likely that if a greater percentage of the population were educated, the non-farm sector would have played a greater role. Women have not directly benefited much by employment in non-farm sectors. However, when men find jobs in non-farm sectors women seem to substitute for them in the agricultural activities. To some extent, this process may have prevented the wage gap across genders from growing. However, the indirect route through which women benefit from non-farm employment suggests that they are more vulnerable than men to the effects of slow agricultural productivity growth. Agricultural productivity increases through technical change remains the principal avenue through which the earnings of women can be increased. To move beyond this point, we need to acquire data on exogenous variables such as total factor productivity in agriculture so that we can examine what level of variation in labor earnings can be explained by the changes in total factor productivity in agriculture. Given that the relative contribution of non-farm sector to labour earnings itself increases with the agricultural productivity growth, it will be interesting to see to what extent the variation in labour earnings can be explained by the variation in agricultural productivity across states. 23

Table 1: Size of NSS Employment Surveys 1999/00 1983 All Rural Urban All Rural Urban # of 701,436 442,025 259,411 714,563 472,573 242,080 Individuals # of 120,578 71,417 49,161 120,897 78,595 42,302 Households # of 10,106 5,999 4,107 10,222 6,961 3,261 Primary Sampling Units 24

Table 2: Assignment of Daily Status Works more than 4 hours Works more than 1 hour and less than 4 hours and is seeking or available for work for more than 1 hour Works more than 1 hour and less than 4 hours and is seeking or available for work for less than 1 hour Works less than 1 hour and is seeking or available for work for 4 hours or more Works less than 1 hour and is seeking or available for work for more than 1 hour but less than 4 hours Employed 1.0 0.5 0.5 0 0 0 Unemployed 0 0.5 0 1 0.5 0 Out of labor force 0 0 0.5 0 0.5 1 Works for less than 1 hour and is seeking or available for work for less than 1 hour 25

Table 3: Employment Structure of Males --- Daily Status 55 th Round 38 th Round Agr Mfg CTT G&P Total Agr Mfg CTT G&P Total AP 0.542 0.102 0.228 0.101 0.972 0.602 0.115 0.164 0.102 0.983 Assam 0.569 0.036 0.196 0.182 0.984 0.705 0.041 0.141 0.104 0.992 Bihar 0.685 0.072 0.151 0.067 0.975 0.712 0.081 0.113 0.075 0.980 Gujarat 0.480 0.175 0.239 0.081 0.975 0.534 0.179 0.152 0.121 0.987 Haryana 0.450 0.173 0.260 0.091 0.975 0.584 0.102 0.161 0.135 0.983 Karnataka 0.578 0.116 0.212 0.070 0.976 0.601 0.117 0.167 0.090 0.974 Kerala 0.325 0.133 0.409 0.089 0.956 0.446 0.153 0.244 0.124 0.967 MP 0.675 0.077 0.155 0.073 0.980 0.694 0.087 0.107 0.081 0.969 MH 0.426 0.164 0.279 0.099 0.969 0.512 0.162 0.195 0.112 0.981 Orissa 0.638 0.082 0.178 0.078 0.976 0.663 0.094 0.113 0.110 0.979 Punjab 0.445 0.158 0.294 0.083 0.979 0.590 0.128 0.165 0.104 0.987 Rajasthan 0.534 0.095 0.258 0.076 0.963 0.659 0.097 0.150 0.081 0.988 Tamil Nadu 0.389 0.205 0.290 0.084 0.969 0.444 0.194 0.212 0.125 0.975 UP 0.574 0.124 0.211 0.077 0.986 0.660 0.107 0.131 0.095 0.992 WB 0.472 0.160 0.267 0.078 0.977 0.505 0.170 0.188 0.115 0.978 All India 0.529 0.125 0.233 0.088 0.976 0.596 0.124 0.157 0.105 0.982 Agr: Agriculture, Mfg: Manufacturing, CTT: Construction, Trade & Hotels, Transport, Storage & Communications, G&P: Government Services, Education, Health, Community Services, Personal Services 26

Table 4: Employment Structure of Females --- Daily Status 55 th Round 38 th Round Agr Mfg CTT G&P Total Agr Mfg CTT G&P Total AP 0.713 0.094 0.089 0.095 0.991 0.716 0.114 0.082 0.080 0.991 Assam 0.686 0.081 0.034 0.191 0.993 0.731 0.065 0.028 0.172 0.995 Bihar 0.783 0.102 0.047 0.061 0.993 0.787 0.094 0.055 0.052 0.988 Gujarat 0.769 0.059 0.081 0.083 0.991 0.809 0.056 0.050 0.083 0.998 Haryana 0.750 0.070 0.063 0.111 0.994 0.797 0.041 0.025 0.134 0.998 Karnataka 0.750 0.114 0.057 0.065 0.986 0.720 0.144 0.070 0.055 0.990 Kerala 0.351 0.278 0.111 0.226 0.966 0.432 0.251 0.066 0.229 0.977 MP 0.837 0.063 0.042 0.050 0.992 0.857 0.058 0.038 0.035 0.988 MH 0.772 0.058 0.064 0.094 0.989 0.773 0.064 0.087 0.069 0.994 Orissa 0.713 0.144 0.071 0.059 0.987 0.691 0.123 0.101 0.076 0.990 Punjab 0.722 0.067 0.039 0.168 0.996 0.572 0.135 0.049 0.232 0.988 Rajasthan 0.853 0.056 0.042 0.041 0.992 0.886 0.043 0.042 0.027 0.998 Tamil Nadu 0.557 0.211 0.109 0.113 0.990 0.603 0.190 0.079 0.122 0.994 UP 0.774 0.100 0.037 0.086 0.996 0.805 0.081 0.044 0.069 0.999 WB 0.388 0.319 0.081 0.206 0.993 0.437 0.207 0.071 0.276 0.991 All India 0.723 0.108 0.066 0.093 0.990 0.744 0.102 0.063 0.084 0.992 Agr: Agriculture, Mfg: Manufacturing, CTT: Construction, Trade & Hotels, Transport, Storage & Communications, G&P: Government Services, Education, Health, Community Services, Personal Services 27

Figure 1: Agricultural Earnings & Labour-Land Ratios: 1983-99 earnings 100 150 200 250 300 350 Punjab Rajasthan Haryana Punjab Haryana Rajasthan Gujarat MH MP Orissa Kerala Assam Karnataka Assam Gujarat Kerala WB UP MH MP Orissa UP Karnataka WB AP AP Bihar Tamil Nadu Tamil Nadu 20000 40000 60000 80000 100000 120000 Labour-Land Ratio Bihar earnings99 earnings83 28

Table 5: Change in Non-Agricultural Employment as a % of Agricultural Employment in 1999 For workers less than or State For all Workers equal to secondary education AP 18.51 11.30 Assam 28.26 22.50 Bihar 11.82 7.20 Gujarat 21.29 14.89 Haryana 21.54 12.01 Karnataka 14.75 7.34 Kerala 44.52 36.36 MP 10.65 4.24 MH 26.53 17.07 Orissa 10.01 6.12 Punjab 30.01 21.81 Rajasthan 18.78 12.28 TN 27.13 15.59 UP 20.23 13.06 WB 24.45 16.13 All India 20.57 12.73 Table 6: Labour Share of Costs: Median Values Crop Share Paddy 28% Wheat 17% Jowar 25% Sugarcane 25% Rapeseed/Mustard 21% Source: Cost of Cultivation of Principal Crops in India, 1996 29

Table 7: Upper Bound Projections of the Contribution of Non-Farm Sector toward Increase in Agricultural Labor Earnings % Change in Earnings for workers with less than or % Change in % equal to % Earnings for % Change in Contribution secondary% Change in Contribution all workers: Earnings: of Non-Farm education: Earnings: of Non-Farm State Observed Projected Sector Observed Projected Sector AP 70.40 22.90 32.54 69.91 14.78 21.14 Assam -7.11 17.71-249.11-9.20 14.39-156.37 Bihar 56.46 14.17 25.10 54.38 8.86 16.29 Gujarat 33.04 20.13 60.93 32.35 14.73 45.53 Haryana 19.70 18.29 92.84 19.56 10.99 56.18 Karnatak 101.07 22.19 21.96 100.23 11.70 11.67 Kerala 66.23 44.68 67.46 65.44 38.34 58.58 MP 57.80 13.01 22.50 55.93 5.41 9.67 MH 68.35 30.52 44.65 63.97 20.56 32.14 Orissa 54.75 12.05 22.02 53.61 7.56 14.11 Punjab 58.26 31.64 54.31 57.49 24.32 42.29 Rajasthan 66.58 22.67 34.05 66.85 15.64 23.40 TN 128.32 42.14 32.84 127.47 26.36 20.68 UP 46.88 21.29 45.42 42.96 14.16 32.97 WB 56.30 26.52 47.10 55.98 18.62 33.26 All India 59.17 23.40 39.54 57.51 15.25 26.52 30

Table 8(a): Sectoral Employment (Males) by Cohort Groups: 1999 Millions of Days Per Week Sectoral Shares in Total 34-42 42-50 50-58 58-66 34-42 42-50 50-58 58-66 Agriculture 157.48 101.01 91.00 66.24 0.49 0.49 0.56 0.71 Mining 3.31 2.41 1.30 0.52 0.01 0.01 0.01 0.01 Manufacturing 40.96 24.64 16.90 8.03 0.13 0.12 0.10 0.09 Construction 19.02 9.07 5.55 2.11 0.06 0.04 0.03 0.02 Trade & Hotels 41.86 24.36 15.96 8.19 0.13 0.12 0.10 0.09 Transport 20.61 11.41 6.95 1.88 0.06 0.06 0.04 0.02 Finance & Real Estate 6.22 4.80 2.68 0.77 0.02 0.02 0.02 0.01 Pub admn & servs 34.66 27.00 21.92 5.81 0.11 0.13 0.14 0.06 Total 324.12 204.70 162.27 93.55 1.00 1.00 1.00 1.00 Table 8(b): Sectoral Employment (Males) by Cohort Groups: 1983 Millions of Days Per Week Sectoral Shares in Total 18-26 26-34 34-42 42-50 18-26 26-34 34-42 42-50 Agriculture 153.86 114.96 109.61 70.88 0.59 0.52 0.55 0.55 Mining 1.78 2.32 2.22 1.30 0.01 0.01 0.01 0.01 Manufacturing 36.18 31.22 25.07 16.72 0.14 0.14 0.12 0.13 Construction 10.46 8.68 7.11 4.08 0.04 0.04 0.04 0.03 Trade & Hotels 25.50 20.30 17.21 10.78 0.10 0.09 0.09 0.08 Transport 10.04 10.95 10.03 5.88 0.04 0.05 0.05 0.05 Finance & Real Estate 1.83 4.51 2.44 1.57 0.01 0.02 0.01 0.01 Pub admn & servs 20.88 28.59 27.35 18.80 0.08 0.13 0.14 0.14 Total 260.53 221.53 201.05 130.00 1.00 1.00 1.00 1.00 31

Table 9(a): Sectoral Employment (Females) by Cohort Groups: 1999 Millions of Days Per Week Sectoral Shares in Total 34-42 42-50 50-58 58-66 34-42 42-50 50-58 58-66 Agriculture 85.20 53.92 41.68 21.80 0.71 0.72 0.75 0.78 Mining 0.78 0.22 0.19 0.06 0.01 0.00 0.00 0.00 Manufacturing 11.70 6.41 3.83 2.32 0.10 0.09 0.07 0.08 Construction 2.54 1.17 0.40 0.17 0.02 0.02 0.01 0.01 Trade & Hotels 5.62 3.80 2.88 1.69 0.05 0.05 0.05 0.06 Transport 0.82 0.28 0.28 0.05 0.01 0.00 0.01 0.00 Finance & Real Estate 0.73 0.38 0.16 0.03 0.01 0.01 0.00 0.00 Pub admn & servs 12.54 8.52 5.91 1.77 0.10 0.11 0.11 0.06 Total 119.91 74.70 55.32 27.90 1.00 1.00 1.00 1.00 Table 9(b): Sectoral Employment (Females) by Cohort Groups: 1983 Millions of Days Per Week Sectoral Shares in Total 18-26 26-34 34-42 42-50 18-26 26-34 34-42 42-50 Agriculture 61.79 54.79 54.40 35.49 0.74 0.73 0.74 0.76 Mining 0.55 0.39 0.52 0.17 0.01 0.01 0.01 0.00 Manufacturing 10.59 7.31 6.62 3.51 0.13 0.10 0.09 0.07 Construction 2.03 1.77 1.63 0.73 0.02 0.02 0.02 0.02 Trade & Hotels 2.15 2.70 3.37 2.17 0.03 0.04 0.05 0.05 Transport 0.37 0.25 0.30 0.13 0.00 0.00 0.00 0.00 Finance & Real Estate 0.27 0.25 0.13 0.08 0.00 0.00 0.00 0.00 Pub admn & servs 5.80 7.59 6.86 4.68 0.07 0.10 0.09 0.10 Total 83.54 75.05 73.83 46.96 1.00 1.00 1.00 1.00 32

Table 10. Average Earnings of Males with no Education for Cohort 34-42 1999 1983 p Sectors w 1i p 1i p 1i w 1i 1i w 1i /W 1 w 0i p 0i p 0i w 0i p 0i w 0i /W 0 Agriculture 210.80 0.68 144.20 0.57 135.80 0.68 92.28 0.56 Mining 444.40 0.02 10.20 0.04 281.08 0.03 8.39 0.05 Manufacturing 369.36 0.07 26.59 0.11 217.86 0.09 18.74 0.11 Construction 295.18 0.11 33.67 0.13 249.00 0.08 20.95 0.13 Trade & Hotels 253.88 0.02 6.00 0.02 150.63 0.02 3.21 0.02 Transport 426.50 0.04 17.34 0.07 215.07 0.04 8.08 0.05 Finance & Real Estate 206.00 0.00 0.46 0.00 212.09 0.00 0.27 0.00 Pub admn & servs 334.95 0.04 13.52 0.05 218.10 0.06 13.15 0.08 Note: W 1 = 252, W 0 = 165.08 Table 11. Average Earnings of Females with no Education for Cohort 34-42 1999 1983 p Sectors w 1i p 1i p 1i w 1i 1i w 1i /W 1 w 0i p 0i p 0i w 0i p 0i w 0i /W 0 Agriculture 147.37 0.80 117.43 0.75 92.72 0.76 70.11 0.73 Mining 238.74 0.01 3.35 0.02 161.83 0.02 2.60 0.03 Manufacturing 179.17 0.04 7.31 0.05 97.05 0.07 7.23 0.08 Construction 192.77 0.05 9.60 0.06 106.60 0.06 6.07 0.06 Trade & Hotels 141.60 0.01 1.00 0.01 109.05 0.00 0.37 0.00 Transport 357.04 0.00 0.82 0.01 117.31 0.00 0.36 0.00 Finance & Real Estate 325.18 0.00 0.03 0.00 103.02 0.00 0.04 0.00 Pub admn & servs 186.53 0.09 16.61 0.11 104.10 0.09 9.31 0.10 Note: W 1 = 156.16, W 0 = 96.09 33

Table 12. Average Earnings of Males with Middle School Education for Cohort 34-42 1999 1983 p Sectors w 1i p 1i p 1i w 1i 1i w 1i /W 1 w 0i p 0i p 0i w 0i p 0i w 0i /W 0 Agriculture 242.67 0.17 41.40 0.08 170.78 0.07 12.05 0.03 Mining 1826.47 0.01 23.78 0.05 346.25 0.01 5.00 0.01 Manufacturing 506.93 0.24 121.19 0.24 343.83 0.21 72.44 0.21 Construction 393.75 0.08 29.87 0.06 298.63 0.02 6.16 0.02 Trade & Hotels 295.94 0.10 29.19 0.06 283.15 0.05 13.03 0.04 Transport 530.28 0.14 72.42 0.15 359.00 0.12 42.85 0.12 Finance & Real Estate 467.17 0.03 13.71 0.03 350.93 0.03 11.35 0.03 Pub admn & servs 705.87 0.24 167.24 0.34 373.90 0.49 181.71 0.53 Note: W 1 = 498.79, W 0 = 344.58 Table 13. Average Earnings of Females with Middle School Education for Cohort 34-42 1999 1983 p Sectors w 1i p 1i p 1i w 1i 1i w 1i /W 1 w 0i p 0i p 0i w 0i p 0i w 0i /W 0 Agriculture 177.00 0.19 33.00 0.08 120.30 0.03 3.99 0.01 Mining 340.57 0.02 6.37 0.02 0.00 0.00 0.00 0.00 Manufacturing 303.63 0.18 55.98 0.14 185.45 0.06 12.03 0.04 Construction 285.51 0.02 5.64 0.01 320.69 0.00 1.39 0.00 Trade & Hotels 214.15 0.04 9.00 0.02 288.26 0.01 4.24 0.02 Transport 600.39 0.04 21.07 0.05 277.56 0.05 15.20 0.05 Finance & Real Estate 358.92 0.01 3.23 0.01 431.44 0.02 8.80 0.03 Pub admn & servs 504.89 0.50 254.76 0.65 292.15 0.81 236.00 0.84 Note: W 1 = 389.05, W 0 = 281.64 34