Improving rural wages in India

Policy, Pannlhg, and Research WORKING PAPgRS Women In Development Population and Human Resources Department The World Bank August 1989 WPS 276 Improving Rural Wages in India Shahidur R. Khandker Do public programs and infrastructure to promote agricultural growth improve real agricultural wages and thus reduce rural poverty? Rural electrification, roads, and banks do - because they increase nonfarm employment. Educational infrastructure, public irrigation, and regulation of markets do not, ,although they raise agricultural output. The Policy, Planning, and Rescarch Cornplex distributes PPR Working Papers to disseminate the findings of work iu progress and to encourage the exchange of ideas amnong Bank staff and all others intcrested in uivelopnent issues. These papers carry the names of the authors, reflect only their views, and should be used and cited accordingly. The findings. intcrpretations, and conclusions are the authors' own. They should not be atuributed to the World Bank, its Board of Directors, its management, or any of its member countries.  Poc,Planning, and Research Women ht Development Do public programs and infrastructure that For example, although educational infra- promote agricultural growth improve real agri- structure, public inigation, and regulation of cultural wages and thus reduce rural poverty? markets raise agricultural output, they depress real agricultural wages because they do not That depends, says Khlandker, basing his increase nonfarm employment conclusions on district-level panel data from India. In contrast, rural electrification, roads, and banks can increase real agricultural wages, Whether public policies increase real because they increase nonfarm employment. agricultural wages depends on whether they promote rural nonfarm employment, to absorb Rural financial institutions and electrifica- the growing labor force. tion reallocate labor from agriculture to rural nonfarm activities, however, while roads pro- mote both farm and nonfarm employment. This paper is a product of the Women in Development Division, Population and Human Resources Department. Copies are available free from the World Bank, 1818 H Street NW, Washington DC 20433. Please contact Belinda Smith, room S9-125, extension 35108 (25 pages with tables). The PPR Working Paper Series disseminaies the findings of work under way in the Bank's Policy, Planning, and Research Complex. An objective of the scries is to get these findings out quickly, even if presentations are less than fully polished. The findings, interpretations, and conclusions in these papers do not necessarily represent official policy of the Bank. Produced at the PPR Dissemination Center  Improving Rural Wages In India by Shahidur RI Khandker Table of Contents I. Introduction 3 II. Model Specification and Estimation Strategy 7 III. Data and Variable Definitions 10 IV. The Results 14 V. Discussion 19 References 24  I. Introduction The introduction of new seed technology in agriculture has enabled India to attain self-sufficiency in basic food grains. Yet, as evidence suggests, poverty remains obstinately high in many parts of rural India (Ahluwalia, 1978; Bardhan, 1985; Vaidyanathan, 1988).1 The continued rural poverty amidst surplus food production is perhaps a result of an inadequate growth in rural employment and income (Mellor, 1988; Mellor and Johnston, 1984). Poverty alleviation thus critically depends on how fast the government can generate productive employment and income for the rural unemployed. One proximate cause of rural poverty is a slow growth in agricultural wage income. Thus, one way to alleviate rural poverty-- National Sample Survey (NSS) of India reports that in 1983 about 40 percent of the rural labor force were wage workers--is to increase real rural wtage. 1/ There are two ways rural poverty has been measured in the literature. One approach quantifies the incidence of poverty (i.e., in terms of the head count ratio, the Sen's index and other measures) based on the mean per capita consumption levels in rural areas and the distribution of the rural population around the mean at different points of time. The other approach looks at the trend of real wage of agricultural workers who are the majority of the rural poors.  The purpose of this paper is to identify public policy and program interventions that can increase rea,. rural wage rate. and hence reduce rural poverty. Rural wages can only increase if the demand for rural labor grows faster than its supply. In other words, rural wages can be affected by changing both the demand for and supply of rural labor. An increase in the demand for rural labor, given its supply, can occur if there is an increased labor demand in either agricultural or nonagricultural activities. Agriculture has limited opportunities to absorb the growing labor force, however. Because of increased agricultural mechanization, the level of farm employment has stagnated or even declined in some parts of India (Bhalla, 1987; Bartsch, 1977). It is, therefore, difficult to raise agricultural real wage unless sufficienit productive employment is generated in the nonfarm sector. Because India's large-scale industrialization policy creates few jobs, employment expansion in the urban sector is not enough to pull labor out of the rural sector (Hellor, 1988).2 Rural nonfarm growth is thus required to generate productive rural employment. The emergence and growth of rural nonfarm pursuits is, however, driven primarily by growth in rural income that leads to higher demand for rural nonfarm goods and services (Binswanger, 1983; Liedholm and Head, 1986). Farm income and wages are important elements of rural income and thus agricultural growth may have an important influence on the growth of rural nonfarm activities and hence demand for labor. According to NSS, the share of nonfarm to total workers 2/ Consequently, rural wages cannot be increased through a reduction in the rural labor force via labor migration.  5 in rural India has risen from 17 (10) percent in 1972 to 23 (13) percent in 1983 among males (females). Agricultural growth, which is about 3.5 percent over this period, perhaps has contributed to this substantial gain in rural nonfarm employment. Does this also mean an increase in real rural wages via induced labor demand? This depends on whether rural labor demand grows faster than the supply. In the long-run rural labor supply can be slowed down if the population growth is reduced. However, labor is mobile and it is difficult to treat rural labor supply as exogenous even in the short-run. For example, government programs and interventions may create job opportunities in a particular area and attract labor from other regions.3 Thus, the prices and infrastructure which affect output supply, employment and wage can also influence the family's labor supply and migration decisions. In other words, labor supply is a household decision influenced by some of the same factors that affect its demand for labor in production. Because labor supply is endogenous, this paper does not attempt to relate agricultural growth to the rural real wages and nonfarm employment (Bardhan, 1985; Khan, (1983); Haggblade, Hazell and Brown, (1988). Although rural real wages and employment (both farm and ncnfarm) tend to be associated with agricultural growth, their simple associations do not tell 3/ Also government may respond to population density for the reason that the marginal cost of providing an infrastructure is lower in a high population density area than a low density one. Because both infrastructure and population influence each other, it is almost impossible to estimate the impact of infrastructure on the population density and hence rural labor supply.  6 us what causes what. More plausibly, they are simultaneously and jointly determined by a number of common exogenous factors influencing farm household's production and consumption decisions. This paper also, unlike other studies (e.g., Binswanger et al., 1987), does not treat population density as an exogenous variable determining output supply. The agroclimatic endowments and infrastructure which affect agricultural output also influence agricultural employment, rural real wages and nonfarm employment. For policy purposes, it is important to know whether the policies that have fostered agricultural growth have had any powerful effect on rural wages and employment. Because the causal factors need not influence the observed outcomes in the same direction, the objective of this paper is to differentiate the factors that promote simultaneous expansion in agricultural production, real wage, farm and nonfarm employment from the factors that exert opposing effects. A central problem of estimating the causal relationships is that public programs and infrastructural investment respond to agricultural opportunities implied by the agroclimatic potentials of an area. Government invests more in a better agroclimatic area where the return to public investment is high. The rural households also respond to better agroclimates by increasing output and thus government programs cannot be considered exogenous to the household's output decisions. Labor demand is determined by agroclimate, infrastructure and the level of output and this makes it difficult to estimate the causal effect of public programs on outcomes such as rural wages and employment. However, these problems can be circumvented if we use a panel dataset. In this paper we use a district-level panel data from 85 districts in India. The paper is structured in the following order. An analytical model based on the theory of farm household production with estimation  technique is outlined in section two. Section three discusses the data and variables used in the paper. The empirical results are discussed in section four. The results are sut.3rized in the concluding part of the paper. 1I1 Model Specification and Estimation Strategy Assume that rural households participate both in farm and nonfarm activities.4 Using the theory of farm household (e.g., Barnum and Squire, 1979), a household's farm and nonfarm output supply or input demand functions can be derived as functions of technology, output and input prices, both the physical and human endowments, and the existence of public institutions and infrastructure. The public institutions and infrastructure determine the "implicit' prices for many goods and services that farm households produce for market and own consumption. Government infrastructure can also directly increase production (either in farm or nonfarm activity) by shifting the production frontier as in the case of irrigation for farm production. Let Fjt be farm production and Njt represent rural (both the farm and nonfarm) employment of district j in period t. Equation (1) relates district-level aggregate farm production and rural employment to the following set of explanatory variables: Fjt, Njt - g(Pjt. Wjt. Pfjt. Bjt. Rjt. /ij; 6) (1) 4I For simplicity, assume that farm and rural nonfarm activities are highly substitutable. However, given household's endowments, a corner solution, i.e., participation only in one activity, is possible for a particular household. Such a distinction is not possible in aggregate district-level data that used in this study.  where Pjt is a vector of J-district's farm and nonfarm output prices in period t; Wjt the wage for hired labor; Pfjt the fertilizer price; Bjt the financial institutions; Rjt the infrastructures acting as shifters of both farm and nonfarm productions; pj is a vector of observable district- specific permanent characteristics; and 6 is the district-specific unobservable characteristics influencing both the farm production and nonfanm activities. Rural households also supply their own family labor taking Wjt as exogenously given; thus the aggregate labor supply in the district J, Sjt, can depend on the saw arguments as in (1): Sjt - h(Pjt, Wjt, Pfjt, Bjt, Rjt. ipj; 6j) (2) We assuma that a multimarket (Hicksian type) competitive equilibrium exists in the labor market, given that active labor market participation by rural households is high in the Indian villages (Rosenzweig, 1978). Moreover, both rural in- and out-migration can act to stablize the labor market equilibrium. Thus, if the demand for rural labor, Njt, is greater (smaller) than the rural supply, Sjt, then in- migration (out-migration) or an increase (decrease) in rural labor supply depresses the market wage until an equilibrium wage is determined. Therefore, the demand for and supply of labor interact to set an equilibrium market wage which is endogeneously determined by (3). W*it - k(Pjt. Pfjt, Bjt, R3t. ji; 6j ) (3) The corresponding equilibrium aggregate crop output supply (F*jt), rural employment (N*it), and rural labor supply (S*jt) can be written as: F*jt - l(Pjt, Pfjt, Bjt, Rjt, /iJ; 6j) (4) N*jt - m(Pjt, Pfjt, Bjt, Rjt, ij; 6j) (5)  9 The relations (3), (4), and (5), respectively, are the estimating equations for the rural wages, agricultural output, and rural employment (farm and nonfarm). The ordinary least squares (OLS) estimation with a cross-section data (i.e., for a given t) is both biased and inconsistent, because the unobserved district-specific characteristics may be correlated with the included right-hand v2riables such as government infrastructure (B). Also, because government infrastructure variables (Ej) are not randomly distributed as often hypothesized (i.e., they are determined partially by the district's permanent factors, ij), the OLS estimates with cross-section data do not tell us whether it is the government infrastructure or the district's permanent attributes that matter most in explaining variation in agricu.ltural output, wage, rural emplo7ment, and rural labor supply. We can circumvent both the endogeneity and unobserved variable problems using a panel dataset with either a fixed or a random effects technique. If the unobserved ability characteristics are time-invariant and specific to each district, then a fixed effects procedure (i.e., dummy variable or differencing out methods) will yield consistent estimates. In contrast, the random effects procedure accounts for the existence of both the time-invariant and time-varying error components. This procedure, however, ignores any correlation between the persistent errors and time- varying observed variables. The fixed effects procedure, on the other hand, does not estimate the influence of the measured but time-invariant variables (e.g., soil moisture capacity) on the dependent variables. We shall use Hausman-Wu specification test to determine whether fixed or random effects technique is appropriate for the given data and present results accordingly.  10 Simultaneity may also arise because of possible endogeneity of district-level agricultural output prices (Pjt) and the fertilizer prices (Pfjt). That is, the district-level agricultural output prices (Pjt) are endogeneously determined by the demand for and supply cf output. We circumvent this output price simultaneity by using the district-level aggregate crop price index based on the international prices of different crops using the district-specific production weights. The aggregate international crop price index is an instrument for the district-level aggregate crop output price. Given that India is a small country in virtually all inter."tiov'al comodity markets, using internarlonal prices completely circumvents the output price endogeneity (Binewanger, Khandker and Rosenzweig, 1988).5 The endogeneity problem for the fertilizer price is minimal. The fertilizer price is a railhead price set by the government at the country level and hence does not respond to district-level demand for fertilizer. 1II. Data and Variable definitions For each district Fj is an aggregate crop index of 20 major crops. No data exist on nonfarm rural wages and the agricultural wage is used as a proxy for rural wages. Wj is the daily real wage rate of male agricultural labor. The wage rate for fieldworker/rloughman is used. Nj measures rural employment. Two stock measures of rural employment are considered which both derive from the decennial population censuses. One is the 5/ Using district-level farm harvest prices, however, may not create endogeneity in the nonfarm employment equation.  agricultural employment measured by the number of male persons employed in agriculture as wage . rkers and the nonfarm employment measured by the total (male and female) persons employed in rural nonfarm activities. In the census employment is measured by occupational status of main workers, i.e., individuals are asked whether they worked in agriculture or non- agriculture for atleast 183 days in the last year prior to the census period. The reason for including female labor in rural nonfarm activities is that women are more active in nonfarm than farm activities. Rural nonfarm activities include activities such as mining and quarrying, manufacturing, processing, servicing, and repairs, construction, trade and commerce, transport, storage and communications, and other services. The crop and wage data are drawn from the data series for 85 districts covering a period of 21 years from 1961 to 1981 used for another study (Binswanger, Khandker and Rosenzweig, 1988). However, because of lack of comparable employment definition used in the census, the employment data from 1961 populatioa census could not be included. The number of districts covered in this paper is 85 which are randomly selected from 13 states of India. The price variables are the aggregate crop price index and fertilizer price index. The price indices are deflated by the consumer price index for agricultural workers using 1975 as the base year. The infrastructure (physical, financial and human) variables include the government irrigation (i.e., area irrigated by canal and tank), the number of regulated markets, the number of rural and semi-urban commercial bank branches, the number of villages electrified, the number of villages with primary school and the road length. All the infrastructure variables are normplized by the district's total geographic area. The persistent time- invariant agroclimatic endowments and locational factors are the length of  rainy seasor, in months, the number of months in a year with excessive rain (where rainfall exceeds potential evapotranspiration), the number of cool months when the mean temperature is below 18 degree farenheit (this is related to the ability to grow wheat), an index of the moisture capacity of the soils in the district, the percentage of district's area under actual or potential irrigation scheme, the percentage of district's area liable to flooding and the district's nearest distance from one of the eight major urban centers in India (i.e., Delhi, Bombey, Calcutta, Hyderabad, Madras, Kanpur, Ahmedabad and Banglore). The only time-varying agroclimatic endowment included in the regression is the district's annual rainfall in millimeter. Annual rainfall is expected to affect the flow outcomes such as agricultural output and wages but not the stock variables such as employment status of a population at different points of time. The mean and standard deviation of the variables are given in table 1. For details on data and variable definition, see Binswanger, Khandker and Rosenzweig, 1988.  13 TABLE 1: Variable Definition and Descriptive Statistics Variable Number of Mean Standard Observations Deviation Dependent Variables Agricultural crop output index 1785 1.192 1.044 Agricultural (male) employment/10 sq. km 170 235.492 196.889 Nonfarm (total) employment/10 sq. km. 170 153.989 206.158 Agricultural real (male) wage, Rs/manday 1785 5.051 2.035 Independent variables Govt. irrigation, '000 ha/10 sq. km. 1785 0.085 0.106 Number of villages with primary schools /10 sq. km 1785 1.140 0.605 Electrified villages, number/10 sq. km 1785 0.688 0.764 Commercial banks, rural branches/10 sq. km 1785 0.069 0.108 Regulated markets, numbers/10 sq. km 1785 0.014 0.022 Total road length, 'OOO km/10 sq. km 1785 4.389 4.277 Aggregate real domestic crop price index 1785 0.968 0.295 Aggregate real international price index 1785 0.687 0.355 Fertilizer price index (real) 1785 3.413 0.505 Annual rainfall (mm) 1785 1138.573 986.503 Length of rainy season in months 85 3.653 1.368 Number of excesa rainy months 85 1.236 1.393 Number of cool months (Temp < 180) 85 0.935 1.313 Percentage of district area liable to flooding 85 1.389 3.531 Irrigation potential, percentage 85 30.001 31.897 Urban distance (km) 85 298.441 152.029 Soil moisture capacity index 85 2.349 1.008  14 IV. The Results The results of joint estimations of agricultural output, rural employment, and real wage are presented in table 2. The Hausman-Wu test suggests that the estimated chi-square statistic is not sufficient to reject the random effect method in favor of fixed effect for explaining variation in growth in the agricultural output, rural employment, real wage and population.6 The results thus indicate that the measured agroclimatic endowments used in the regression represent a sufficiently precise quantitative characterisation of the agroclimatic potential of a district. An increase in agricultural output price increases crop output and both farm and nonfarm employment. Whether this increases rural wages is not clear, however. Thus, the idea that increasing farm harvest prices can benefit the rural poors more than it hurts them by raising the food prices (Lipton, 1984; Tyagi, 1979) is not evident in this dataset. The response of an increase in agricultural output price is the highest for rural nonfarm employment with an elasticity of 0.20 followed by agricultural output with an elasticity of 0.19 and agricultural employment with an elasticity of 0.15. A 10 percent increase in the price of fertilizer decreases agricultural output by 21 percent because of a negative profit effect on output and income. The same percent increase in the fertilizer price 6/ For a small number of time periods and large number of cross-section units, it is better not to reject the random effects model unless the estimated chi-square statistics is sufficiently higher than the critical level (Maddala, 1987).  15 increases rural wages by 13 percent, perhaps implying increased demand for agricultural labor to substitute fertilizer in farm production. Government investment on roads has a positive effect on crop output, rural nonfarm employment and agricultural real wages. The results suggest that better roads increase both the farm and nonfarm productions and hence agricultural real wages because of induced labor demand. The response of an increase in road investment is the highest for rural nonfarm employment with an elasticity of 0.2 followed by agricultural employment with an elasticity of 0.07, agricultural output with an elasticity of 0.06 and rural wages with an elasticity of 0.04. Government irrigation increases agricultural output and yet reduces rural wages. Irrigation attracts more labor than it perhaps provides jobs and thus depresses rural wages. A 10 percent increase in government irrigation increases agricultural output by about 6 percent, a significant effect of public irrigation on the private output supply. It reduces agricultural real wage by about 4 percent.  16 Table 2: Effect. of Agroeclmtic Endowments, Innfrstructura, Banks and Pricee on Agrcultural Output, Wage, Rural Employmnt and Population Explanatory Variable Aggregate Agricultural Nonfarm Agricultural crop output employment employment real wage Aggregate real crop price (laggod)5 0.194 0.160 0.204 0.002 (9.161)* (3.244)- (2.186) i (0.144) Real fertilizer pricea -0.214 0.244 0.012 0.185 (-3.989)0 (0.716) (0.032) (8.23So Road 0.06 0.074 0.241 0.03C (2.017). (0.782) (2.559). (1.697). Government Irrigationa 0.06S 0.014 0.070 -0.038 (1.738)* (0.220) (1.071) (-1.658) Regulated marketes 0.202 0.06S -0.048 -0.021 (12.804)* (1.637)* (-1.282) (-1.746). Comme rcial bankeS 0.039 -0.06 0.292 0.044 (8.642)* (-2.694)o (10.942)* (6.532)e Primary echoolo' 0.177 0.172 -0.527 -0.288 (2.498)* (1.214) (-8.572). (-4.433). Rural electrificatlon& 0.112 -0.061 0.112 0.061 (6.399). (-1.686). (8.084). (4.561). Year -0.019 1.562 0.279 0.045 (-6.659)e (1.894) (O.81) (4.904)o Rainfall x1O0 0.071 0.186 (3.293). (1.968)e Soil moleture capacity -0.079 -12.760 -9.741 -0.089 (-1.059) (-0.542) (-0.652) (-0.480) Urban dltance -0.0004 0.156 0.025 0.001 (-0.548) (0.941) (0.282) (0.922) Length of ralny season 0.142 5U.896 29.818 -0.07 (1.752). (2.095)* (1.790)* (-0.322) Excess rain months 0.087 16.583 86.769 0.042 (0.478) (0.768) (2.831)e (0.218) Cool winter months 0.058 -39.866 18.447 0.364 (0.886) (-2.024). (1.072) (2.234). Flood potential -0.082 1.265 -8.60 -0.057 (-1.815)