对印度死亡率降低的评价：补偿工资微分研究

WPS-2078 Valuing Mortality Reductions in India A Study of Compensating Wage Differentials by Nathalie B. Simon, Maureen L. Cropper Anna Alberini, and Seema Arora January 1999 Nathalie B. Simon is an economist at USEPA. Maureen L. Cropper is a principal economist at the World Bank (DECRG). Anna Alberini is an assistant professor at the University of Colorado and Seema Arora is an assistant professor at Vanderbilt University. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent nor do they represent those of USEPA. FILE COPY  Valuing Mortality Reductions in India A Study of Compensating Wage Differentials by Nathalie B. Simon, Maureen L. Cropper Anna Alberini, and Seema Arora I. Introduction Conducting benefit-cost analyses of health and safety regulations requires placing a dollar value on reductions in health risks, including risk of death. In the United States, mortality risks are often valued using compensating wage differentials (Viscusi, 1992 and 1993; Viscusi and Moore, 1989; Moore and Viscusi, 1988). These differentials measure what a worker would have to be paid to accept a small increase in his risk of dying, or, equivalently, what the worker would pay to achieve a small reduction in his risk of death. These values, estimated from observed labor market data and converted to Values of a Statistical Life (VSL), are used to value reductions in risk of death achieved by industrial safety programs or environmental health programs (USEPA, 1997; Bayless, 1982).1 Although there is an extensive literature on compensating wage differentials in the U.S., very few studies of compensating wage differentials exist in developing countries. In the absence of these studies, policy makers often value improvements in life ' The Value of a Statistical Life is the value of a risk reduction divided by the size of the risk reduction: VSL Value..of.-Ar Ar 1  expectancy using foregone earnings. That is, the value of an improvement in life expectancy is measured by the income that would be earned during the individual's additional productive lifetime. It is well known that valuing mortality risks using earnings--the so-called human capital approach--is likely to understate what people themselves would pay for small risk changes. Indeed, U.S. data suggest that the amount people would pay for risk reductions based on compensating wage differentials is from 8 to 23 times the value of the risk reductions as estimated by the human capital approach (Viscusi, 1993). One way to estimate a VSL for developing countries without conducting original studies is to extrapolate VSL estimates to these countries from the U.S. For example, one could inflate the foregone earnings of an Indian worker by the ratio of the VSL to foregone earnings computed from a study in the U.S. This assumes, however, identical attitudes toward risk in the two countries and an income elasticity of willingness to pay (WTP) for risk reductions equal to one.2 Without independent studies of the VSL in developing countries, such assumptions are difficult to justify. We attempt to shed light on this issue by conducting a compensating wage study in one developing country-India-to obtain estimates of the VSL that reflect Indian risk preferences. We then compare the ratio of the VSL to the present value of foregone earnings in India with ratios obtained in similar studies conducted in the U.S. and 2 These points are discussed more fully in Section II. 2  elsewhere. This allows us to evaluate the benefits-transfer methodology discussed above, using India as an example. The remainder of the paper proceeds as follows. In section II, we discuss the theory of compensating wage differentials and the theoretical basis for the benefits- transfer methodology discussed above. Section III reviews selected studies in the compensating wage literature, including those for developing countries. Section IV describes the methodology and data we use to estimate the Indian compensating wage differentials. Section V presents our results, compares them to those obtained in the U.S. and examines the implications for benefits transfer. Section VI concludes. II. Compensating Wage Differentials and Benefits-Transfer A. An Hedonic Model of the Labor Market The basic idea behind compensating wage differentials is that jobs can be characterized by various attributes, including risk of accidental death. Workers are described by the amount they require as compensation for different risk levels, while firms are characterized by the amounts they are willing to offer workers to accept different risk levels. The matching of wage offers and acceptances determines the hedonic wage equation, which describes the compensation received for bearing risk in market equilibrium. One way to express the individual's willingness to substitute risk for income in the labor market is to ask how much compensation he would require to work at various 3  risk levels, holding utility constant. This compensation is the amount C that is given implicitly by (1- p)(1 - r)U(C + I) = k (1) where r is the risk of death on the job, p is the risk of dying from all other causes and I represents non-wage income. The compensation function C(r) is pictured in Figure 1 for different workers. The worker's choice of risk level, r, occurs where a marginal change in required compensation, C'(r), equals a marginal change in the wage offered in market equilibrium, w'(r), or, equivalently, where the compensation function is tangent to the hedonic wage equation, w(r). From the firm's perspective, wage offers that keep profits constant at various risk levels are given by offer curves PP in Figure 1. Equilibrium in the labor market is given by the locus of tangency points between various required compensation and offer curves. This locus is the hedonic wage function, and its derivative with respect to risk of death measures the value of a small change in risk to the worker: dw (1- p)U(w + I) VSL (2) dr (1- p)(l - r)U'(w+ I) The rate at which a worker is willing to substitute income for risk is his expected utility if he survives risk of death on the job, (1-p)U'(w+I), divided by his expected marginal utility of income, (l-p)(1-r)U'(w+I). VSL, as we label the relationship, is increasing in risk, indicating that, ceteris paribus, individuals facing higher baseline risk should be willing to pay more for a change in risk of a given magnitude. Representing 4  Figure 1: Equilibrium in an Hedonic Labor Market Wage w(r) C2( r ) P2 CI(r) P2 Pi P1 Risk Level (r) 5  income by Y _ w + I, we examine the relationship between the value of a risk change (VSL) and (foregone) income, which forms the basis for benefits transfer. B. Benefits-Transfer Using Ratios of VSL to Foregone Earnings If we assume that individuals in two different countries, say A and B, have the same preferences (i.e., have the same utility function), then we can assume that, ceteris paribus, the ratio of VSL to foregone earnings is the same in both A and B, or: VSLA VSLB (3) yA yB where Y equals the present value of foregone earnings. In this case, transferring estimates of VSL from country A to country B is a simple matter. One can either multiply yB by the ratio (VSLA/YA) or multiply VSLA by the ratio (yB/yA).3 This benefits transfer methodology is valid, however, only if the ratio (VSL/Y) is independent of income. This is equivalent to assuming that the VSL increases in strict proportion to income, i.e., that the income elasticity of the VSL is equal to 1. If, on the other hand, the VSL increases less than in proportion to income (if the income elasticity is less than one), transferring VSL estimates from a richer to a poorer country will under- estimate the VSL for the poorer country. Of course, the above discussion assumes that individuals' preferences, and hence the utility functions faced by individuals in both countries, are identical. If they were to 3 The former method, multiplying VSLB by the ratio of VSLA/YA, avoids the use of exchange rates and thus provides more stable estimates. 6  vary across countries, conditions for benefits transfer would become even more restrictive. This underscores the importance of conducting compensating wage studies to estimate the VSL in developing countries. III. Review of Labor Market Studies A. Compensating Wage Studies in Developed Countries Although the notion that workers will be compensated for unpleasant conditions in the workplace dates back to at least Adam Smith (1776), the modem theory of compensating wage differentials is usually attributed to Richard Thaler and Sherwin Rosen (1976). Thaler and Rosen suggested that the labor market could be viewed as an example of an hedonic market-a market in which the price of a product (in this case, a job) varies with its attributes. In the two decades since Thaler and Rosen's pioneering work, there have been literally dozens of studies that have estimated hedonic wage equations. Hedonic wage equations have been estimated for various segments of the U.S. labor market, including unionized industries, blue-collar workers, and males. Similar studies have been conducted in the United Kingdom (Marin and Psacharopoulous, 1982), Australia, Japan (Knieser and Leeth, 1991), and Canada (Cousineau et al., 1992). While the earliest of these studies used average wage data across industries (Smith, 1974), later contributions focused on data for the individual worker. In a typical study the wage received by worker i is explained as a function of his productivity (measured by education, experience, job tenure) and job attributes. Whenever possible, 7  the data are measured for the individual; however, risk of accidental death or injury is usually measured for the industry in which the individual works.4 Typically, the risk data used in these studies are reported for two-digit industrial classifications, although recent studies in the U.S. use risk data reported at the one-digit industry level by state (Viscusi and Moore, 1989; Moore and Viscusi, 1988, 1990a and 1990b; Knieser and Leeth, 1991). In addition to including risk of accidental death, it is important that the hedonic wage equation include job attributes that might be correlated with risk of death on the job. This includes risk of non-fatal injury, as well as dummy variables for the worker's occupation and the industry, to capture non-pecuniary job attributes that are difficult to measure directly. Several studies have shown, however, that, even after controlling for individual and job characteristics including risk of death and injury, equally skilled workers are able to command higher wages in some industries than in others (Krueger and Summers, 1988; Katz, 1986). The explanation generally offered for these persistent inter-industry differentials is that some firms are simply willing to pay higher wages-perhaps to avoid strikes, labor turn-over, etc.-and find that paying these higher wages is consistent with maximizing profit. This underscores the importance of controlling for inter-industry wage differentials in measuring the premium attached to fatal and nonfatal job risks. 4Several studies (Gegax, De Haan, and Schulze, 1991; Gerking and De Haan, and Schulze, 1988; Shanmugam, 1997) have used the individual's subjective estimate of his or her own risk as the measure of risk. The correlation between perceived and actual risk, however, was generally low in spite of the fact that risk so defined was a significant predictor of wage. 8  B. Compensating Wage Studies in Developing Countries In spite of the large number of compensating wage studies completed in developed countries, very few have been completed in developing countries. However, in a recent study for Taiwan, a newly industrialized economy, Liu et al. (1997) find evidence of compensating wage differentials using data from labor market surveys for years 1982 to 1986. The implied VSL stemming from their work ranges between US $413,000 and US $461,000 with a ratio of VSL to the present value of foregone earnings between 7 and 8. Only one other study that we know of has been conducted in a developing country-Shanmugam's work in India (1997). Unlike our study, however, Shanmugam's study focuses on a single metropolitan area in India--the Madras District in the state of Tamil Nadu--rather than looking at the country as a whole. Using 522 observations collected from blue-collar, male workers including wage, worker characteristics and perceptions of job risk, Shanmugam finds evidence of compensating wage differentials using injury data collected at the two-digit level (Indian Industrial Classification) as the risk measure. The implied value of a statistical life comes to Rs. 12.084 million, or approximately US $400,000. Looking at the ratio of VSL to the present value of foregone earnings, however, Shanmugam's figures seem exceptionally high. The ratio of his VSL to foregone earnings is 73, 10 times higher than that obtained by Liu et al. for Taiwan and between 5 to 13 times higher than that obtained by Moore and Viscusi for the U.S. (1988). It could be that by focusing on only one metropolitan area, his estimate of VSL, and hence the 9  ratio of VSL to foregone earnings, is not representative of that for the country as a whole. We remedy this in our study by using labor market data for the entire country. IV. Methodology We estimate an hedonic wage equation for Indian manufacturing industry using grouped data for 1,454 occupation/industry cells. The hedonic wage equation takes the following form, Inwe = ao +ajr + ,q5Di +Xii ++FIZ (4) i=1 where wy represents the average daily wage in occupation i and industry j, rj is the risk of fatal injury in industry j, and the {Di} are occupation dummies. The vector Xy includes characteristics that vary by occupation and industry, such as the proportion of men in the occupation/industry cell, while Zj includes industry-specific data such as average firm size, region in which the industry is concentrated, and risk of non-fatal injury. Since we use grouped data, we estimate the model using weighted least squares, where the weights are the number of workers in a given occupation-industry cell. The data used to estimate the above equation are described in more detail below. A. Data Description Data on the average daily wage earned by workers in occupation i and industry j are from the Occupational Wage Survey (OWS), conducted by the Indian Labour Bureau. The OWS is administered periodically to a random sample of registered factories in 10  selected industries.5 Industries are included in the OWS if they belong to the organized (i.e. unionized) sector, employ a "large" number of individuals and if they are considered to be "important" to the Indian Industrial Economy (OWS, 1994). Given that unions play a potentially large role in informing their members of job-related risks, it makes sense to focus on this sector of the Indian labor force. The Fourth Round of the OWS, conducted between 1985 and 1991, provides the most recent, completed survey available. It compiles information for a total of 53 three-digit industries, listed in Table 1. While the OWS provides information for some plantation industries and mining, we limit the analysis to manufacturing only. The OWS reports the average daily wage for full-time, manual workers, by occupation, for the industries listed in Table 1.6 In addition to wage information, the OWS provides other pertinent information at the occupation level within each industry, including a detailed description of the work performed, the number of workers employed, the proportion of men, the number of time-rated (vs. piece-rated) workers, cost-of-living allowance, over-time allowance and shift allowance. Table 2 lists the occupations that occur most frequently in the data. The OWS also provides some information for each All factories employing 20 or more persons (10 or more if power is used) are required to register with the relevant state government. In addition, each state has the authority to extend the registration requirement to any factory regardless of size provided that, if the operation is family-owned, at least one employee is hired from outside the family. 6 Because the OWS data took several years to collect, all wage and other monetary data are reported in constant (1990) rupees. 11  Table 1: Manufacturing Industries Included in Fourth Round of the Occupational Wage Survey 2-digit IIC 3-digit IIC Industry Name 201 Manufacture of Milk and Ice Cream Powder, Condensed and Bottled Milk, and Baby Milk Foods 20&21: Food and 206 Manufacture and Refining of Sugar (Vacuum Pan Sugar Factories) Food Products 207 Manufacture of Khandsari Sugar 210 Manufacture of Hydrogenated Oils, Vanaspati Ghee, etc. 212 Tea Processing 214 Cashewnut Shelling, Processing and Packing 22: Beverages, 224 Soft Drinks and Carbonated Water Industries Tobacco and Related 227 Manufacture of Cigarette and Cigarette Tobacco Products 23: Cotton Textiles 231 Cotton Spinning, Weaving, Finishing 24: Wool, Silk and 241 Wool Spinning, Weaving and Finishing in Mills Man-Made Fibre 245 Spinning, Weaving and Finishing of Silk Textiles Textiles 247 Spinning, Weaving & Finishing of Other Textiles, Synthetic Fibres 25: Jute 251 Jute and Mesta Spinning and Weaving 26: Textile Products 264 Manufacture of All Types of Textile Garments including Wearing Apparel 28: Paper & Paper 280 Manufacture of Paper, Newsprint and Packing Paper (Machine Made) Products & Printing, 284/285 Printing and Publishing of Newspapers, Periodicals, Books, Journals, Maps, etc. Publishing, etc. 29: Leather and 291 Manufacture of Footwear (excl. Repair) Except Vulcanized, Moulded Rubber, Leather Products Plastic Footwear 30: Basic Chemicals 300 Manufacture of Tyres and Tubes &Chemical Products 304 Petroleum Refineries 310 Manufacture of Basic Industrial Organic, Inorganic Chemicals and Gases 31: Rubber, Plastic, 311 Manufacture of Fertilizers (Inorganic, Organic and Mixed) Petroleum and Coal 313 Manufacture of Drugs and Medicines 314 Manufacture of Toilet Soap, Washing Soap and Soap Powder 317 Manufacture of Matches 32: Non-Metallic 321 Manufacture of Glass and Glass Products Mineral Products 324 Manufacture of Cement 33: Basic Metal and 330 Iron and Steel Industries Alloys 331 Manufacture of Castings and Forgings (Ferrous) 335 Aluminum Manufacturing Industries 350 Manufacture of Agricultural Machinery and Equipment and Parts 352 Manufacture of Prime Movers, Boilers and Steam Generating Plants and Parts 353 Manufacture of Textile Machinery and Jute Machinery 35/36: Machinery 355 Manufacture of Refrigerators and Air Conditioners and Equipment Other 357 Manufacture of Machine Tools, their Parts and Accessories than Transport 360 Manufacture of Electrical Industrial Machinery and Apparatus and Parts Equipment 363 Manufacture of Electrical Apparatus, Appliances and Other Parts Exel. Repair 364 Manufacture of Television Sets and Teleprinters 366 Manufacture of Electronic Computers, Components and Control Instruments and Accessories 370 Ship Building and Repairing 371/372 Manufacture of Locomotives, Railway Wagons and Coaches and Parts 374 Manufacture of Motor Vehicles and Parts Equipment and Parts 375 Manufacture of Motorcycles and Scooters and Parts 376 Manufacture of Bicycles, Cycle Rickshaws and Parts 377 Manufacture of Aircrafts and Parts 38: Other 382 Manufacture of Watches and Clocks IIC=lndian Industrial Classification 12  Table 2: Occupations Appearing in 10 or more OWS Industries Occupation Name Number of Industries Watchman 45 Helper 44 Sweeper 43 Coolie 41 Driver 41 Electrician 41 Fitter 41 Supervisor 40 Carpenter 32 Checker 31 Welder 30 Packer 29 Foreman 28 Machine Operator 28 Turner 28 Mechanic 26 Painter 21 Blacksmith 19 Maistry 18 Grinder 17 Oilman 16 Boiler Attendant 15 Driller 15 Moulder 13 Pressman 13 Fireman 12 Mason 12 Mate 12 Assembler 11 Charge Hand 11 Furnace Man 11 Loader/Unloader 11 Wire Man 11 Crane Driver 10 13  three-digit industry, including average firm size and the concentration of the industry within geographic regions. We supplement the OWS data with data collected and reported in the Annual Survey of Industries (ASI). These data include value added per worker, number of employees in government-operated (vs. privately-owned) factories and the turnover rate in each industry. These data are reported at the two-digit industry level, except for value added per worker, which is reported at the three-digit level. To measure on-the-job risk, we use the rate of fatal injuries in the industry and the rate of non-fatal injuries in the industry, obtained from various volumes of the Indian Labour Yearbook. Descriptive statistics and definitions for all of the variables used are provided in Table 3. The average annual wage rate for Indian manufacturing workers is Rs. 20,054 (1990 Rupees).7 In general, risk of death on the job is somewhat higher in India than in the U.S. In India, approximately 15 workers per 100,000 die on the job compared to only 8 per 100,000 in the U.S. (Moore and Viscusi, 1988).8 B. Regression Analysis Using the data described above, we estimated several different specifications of the hedonic wage equation. These vary in the number of explanatory variables included in the model, and in the definition of the risk variables. 7 The average annual wage was calculated assuming a 300-day work-year. 8 Indian job-related fatality rates are reported per 10,000 man-days. Rates per 100,000 worker are calculated assuming a 300-day work-year. 14  Table 3: Descriptive Statistics Variable Level of Standard Name Variable Definition Observation Mean Deviation Wage Annual Wage* (1990 Rupees) 3-digit industry/ 20054.01 8815.48 occupation Fatal Number of fatal injuries/ 100,000 2-digit industry 0.149 0.086 workers, from previous period Nonfatal Number of non-fatal injuries/ 2-digit industry 38.942 42.493 100,000 workers, from previous period Workers Number of workers in occupation 3-digit industry/ 3505.37 9827.39 occupation Time Proportion of time-rated workers 3-digit industry/ 89.258 28.461 occupation Overtime =1 if overtime is paid 3-digit industry/ 0.711 0.453 occupation Shift =1 if shift allowance paid 3-digit industry/ 0.569 0.495 occupation Icode25 =1 if iic=25 (Jute textiles) 2-digit industry 0.030 0.169 Icode29 =1 if iic=29 (Leather & Fur) 2-digit industry 0.015 0.122 Icode30 =1 if iic=30 2-digit industry 0.085 0.278 (Plastic & Petroleum) Icode36 =1 if iic=36 (Electric Machinery) 2-digit industry 0.119 0.324 Icode37 =1 if iic=37 2-digit industry 0.140 0.347 (Transport Equipment) Men proportion of men 3-digit industry/ 96.083 13.462 occupation Unitsize average unit size ('00s) 3-digit industry 2.289 4.609 Value average value added per 1000 3-digit industry 57.190 64.231 workers, in 1990 Rupees South =1 if share of workers 3-digit industry 0.117 0.321 in south > 40 % Nwest =1 if share of workers 3-digit industry 0.021 0.142 in northwest > 40% East =1 if share of workers 3-digit industry 0.075 0.263 in east> 40 % West =1 if share of workers 3-digit industry 0.147 0.354 in west > 40 % Public Proportion of workers in 2-digit industry 0.274 0.146 government-operated industries Dispute average length of industrial 2-digit industry 1.488 2.320 disputes per 1,000 mandays Quit average turnover rate 2-digit industry 30.289 11.174 Notes: Annual wage includes basic wage, dearness allowance, shift allowance, overtime, housing allowance if applicable. 15  The core model controls for occupation, the proportion of men and the proportion of time-rated workers employed in an occupation/industry cell. It also includes indicator variables to control for shift allowances and overtime allowances, and includes average value added per worker. The 455 occupation dummies control for differences in required skill and education levels for each job. The proportion of men attempts to control for wage differentials between male and female workers: Occupations employing a higher proportion of men presumably will pay higher wages all else equal. The presence of shift allowances and overtime allowances are used as indicators of the pleasantness of the job. If shift allowances and overtime allowances are paid, this may indicate that the job requires long hours of work at less convenient and desirable times of the day. The average value added per worker, on the other hand, acts as a proxy for worker productivity. The model also controls for the average size of establishments in the industry. The second model, in addition to the variables listed above, controls for the region in which the industry is concentrated, to account for any differences in cost of living or other geographical factors that could influence wages.9 The indicator variable for a region equals one if 40 percent or more of the workers employed in an industry are located in the region in question. The third and final specification incorporates several other variables reported at the two-digit industrial classification level including the proportion of workers employed 9 The regions were defined according to specifications provided in Hanson and Lieberman's India:Poverty, Employment and Social Services, 1989. 16  in government-operated factories (vs. privately owned and operated), the average length of industrial disputes in man-days, and the average turnover rate in the industry. While one would expect that workers employed in government-operated factories enjoy more job security, workers employed in such firms may in return earn lower wages than those employed in private companies for the same work. The average length of labor disputes in an industry acts as an imperfect proxy for the strength of the unions. Reliable information on union membership was not available. Finally, we added the average turnover rate to the model to control for the fact that new hires at a factory are likely to earn less than workers with more tenure. The higher the rate of turnover, the lower the average tenure and the lower the average wage. In each of the specifications described above, we have attempted to control for inter-industry differentials by incorporating several industry dummy variables. However, given that the injury data are available only at the two-digit level and that the scope of this analysis extends to manufacturing industries only, it is impossible to control for all industries. In order to select dummy variables for the most influential industries, we first regressed the wage on the full set of industry dummy variables as well as other occupation characteristics, using weighted least squares. Those industry dummies that were significant at the 99.9 percent level were then incorporated into the hedonic wage equation. The results for this initial regression are reported in Table 4. The three specifications of the hedonic wage equation described above were estimated for each of two measures of fatal and non-fatal injury risk. The first uses the 17  Table 4: WLS+ Regression Results for Selection of Industry Dummies (Dependent Variable: Log Wage; Occupation Dummies Absorbed) Variable Name Coefficient St. Error Pr > ITI FOOD PRODUCTS -0.004 0.099 0.9709 TOBACCO & TOBACCO 0.354** 0.159 0.0260 PRODUCTS COTTON TEXTILES 0.211** 0.104 0.0424 WOOL, SILK & SYNTHETIC 0.255** 0.097 0.0088 TEXTILES JUTE, HEMP & MESTA 8.380** 1.908 0.0001 TEXTILES TEXTILE PRODUCTS -0.094 0.111 0.3977 PAPER & PAPER PRODUCTS 0.216** 0.104 0.0381 LEATHER & FUR PRODUCTS 0.446** 0.122 0.0003 RUBBER, PLASTIC, 0.442** 0.096 0.0001 PETROLEUM & COAL CHEMICAL & CHEMICAL 0.223 0.125 0.0754 PRODUCTS NON-METALLIC MINERAL 0.255** 0.103 0.0132 PRODUCTS BASIC METAL & ALLOYS 0.160 0.095 0.0927 INDUSTRIES MACHINERY & MACHINE 0.139 0.101 0.1685 TOOLS ELECTRICAL MACHINERY & 0.421** 0.094 0.0001 APPLIANCES TRANSPORT EQUIPMENT & 0.406** 0.094 0.0001 PARTS MEN 0.005** 0.001 0.0001 UNITSIZE 0.1412** 0.028 0.0001 UNITSIZE*UNITSIZE -0.016** 0.004 0.0001 AVGVALD 0.003** 0.0004 0.0001 CONSTANT 8.724** 0.116 0.0001 Notes: +Weight=number of workers employed in occupation-industry cell. ** Indicates significance at 95% level of confidence. 18  fatal and non-fatal injury rates in the year preceding the collection of wage data in the OWS. The second uses the average of the injury rates from the five years preceding the OWS. V. Discussion A. Regression Results The results for all six model specifications are provided in Table 5. The signs of the coefficients of most variables generally agree with what has been reported in the literature. In all six specifications, occupations with a higher proportion of men pay more than those employing more women. Workers in large firms earn more than those working in small firms, but at a decreasing rate. Workers are also rewarded for higher productivity as evidenced by the positive (and statistically significant) coefficient on the value-added variable. Wages are higher in occupations that offer shift allowances and/or overtime allowances, perhaps to compensate workers for unpleasant working conditions. Region also seems to play a significant role in determining the wage although not as we had first anticipated. The region concentration variables were introduced into the equation primarily to account for cost-of-living differences across the country. Instead, it appears that they are capturing a different phenomenon. By construction, the regional concentration variable equals one if 40 percent or more of the employees in the industry are located in the region in question. The negative coefficients we obtained for every region suggest that, when an industry is concentrated in a region, firms in the industry 19  Table 5: WLS+ Regression Results for Six Model Speciflcations (Dependent Variable: Log Annual Wage; Standard Errors in Parentheses) Variable Risk From Previous Year Five-year Average Risk Name Model 1 Model 2 Model 3 Model I Model 2 Model 3 Constant 8.410** 8.468** 8.269** 8.480** 8.563** 8.262** ________________ (0.634) (0.568) (0.574) (0.640) (0.581) (0.579) Fatal 0.430** 0.684** 0.747** 0.318 0.324 0.601* (0.1300) (0.124) (0.145) (0.275) (0.255) (0.330) Nonfatal 0.0023** 0.002** 0.004** 0.002** 0.001* * 0.006** (0.0004) (0.0004) (0.001) (0.0004) (0.0004) (0.001) Time 0.001** -0.0004 -0.001 0.001** -0.0003 -0.001 (0.0005) (0.0004) (0.0004) (0.0005) (0.0005) (0.0005) Overtime 0.105** 0.059** 0.063** 0.084** 0.066** 0.071** (0.024) (0.025) (0.025) (0.025) (0.026) (0.025) Shift 0.033 0.044** 0.040** 0.045** 0.049** 0.044** (0.022) (0.020) (0.020) (0.022) (0.021) (0.021) Icode25 6.861** 9.312** 10.311** 4.928** 6.928** 9.503** (Jute) (1.535) (1.485) (1.705) (1.514) (1.522) (1.709) Icode29 0.385** 0.356** 0.326** 0.361** 0.316** 0.314** (Leather) (0.083) (0.076) (0.079) (0.086) (0.079) (0.079) Icode30 0.268** 0.361** 0.394** 0.246** 0.326** 0.392** (Rubber) (0.034) (0.033) (0.038) ((0.034) .034) (0.039) Icode36 0.329** 0.393** 0.429** 0.282** 0.307** 0.362** (Elec. Mach.) (0.037) (0.035) (0.038) (0.038) (0.037) (0.040) Icode37 0.264** 0.358** 0.378** 0.248** 0.333** 0.242** (Transport) (0.026) (0.025) (0.059) (0.036) (0.032) (0.085) Men 0.006** 0.007** 0.008** 0.006** 0.008** 0.007** _____________ (0.001) (0.001 ) (0.001) (0.001) (0.001) (0.001) Unitsize 0.118** 0.135** 0.149** 0.081** 0.090** 0.126** _______________ (0.021) (0.020) (0.025) (0.020) (0.021) (0.025) (Unitsize)^2 -0.014** -0.018** -0.020** -0.010** -0.013** -0.017** ________________ (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) Value 0.004** 0.003** 0.002** 0.004** 0.003** 0.003** _______________ (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) (0.0003) South -0.171** -0.202** -0.137** -0.180** (0.030) (0.031) (0.030) (0.031) Nwest -0.736** -0.757** -0.734** -0.751** (_________ 0.052) (0.052) (0.053) (0.053) East -0.169** -0.275** -0.089** -0.214** (__________ 0.034) (0.045) (0.033) (0.045) West -0.034 -0.047 -0.025 -0.033 (0.023) (0.026) (0.024) (0.026) Public -0.223 -0.162 (0.134) (0.150) Dispute -0.022 -0.077** (0.014) (0.020) Quit 0.001 0.001 ________________ ___ _______ __________ _ (0.001) _(0.002) Notes: +Weight=number of workers in occupation-industry cell. **Indicates significance at the 95% level of confidence and * at the 90% level. Coefficients for Occupation Dummies omitted. 20  have monopsony power or are selecting locations strategically to take advantage of low labor costs. As expected, government ownership of firms reduces wages, as do longer labor disputes. The effect of the average turnover rate, on the other hand, is positive, an unexpected sign, although not statistically significant. Judging from the significance of the industry dummy variables, inter-industry wage differentials appear to be an important determinant of wages as well. Finally, turning to the effects of the fatal and non-fatal injury rates, risk of injury and risk of death on the job positively influence the wage. This holds true in all specifications of the model; however, only the coefficients using the previous period's injury and mortality rates as the risk measures produce statistically significant results. This lack of statistical significance is not surprising, however, since averaging the risk variables over five years reduces the variation in the risk measure across industries and thus increases the standard errors of the risk coefficients. The fact that the magnitudes of the coefficients are similar across definitions of risk, however, is indicative of the robustness of the results. B. Value of Life Estimates Since the compensating wage differentials obtained from studies like this are used to value improvements in risk of death and/or injury, it is interesting to convert these estimates to Values of a Statistical Life (VSL) and Values of a Statistical Injury (VSI) for comparison to other studies. According to our models, the VSL, when evaluated at the 21  mean wage, ranges between Rs. 6,417,341 and Rs. 15,040,642 (1990 Rupees).10 The VSI, on the other hand, ranges between Rs. 20,054 and Rs. 120,325 (1990 Rupees). At current exchange rates ($1 US = 42 Rs.) the VSL for India ranges between US $153,000 and US $358,000, while the VSI ranges between US $477 and US $2,870. In judging whether these results are reasonable, it is important that the Indian VSL values be compared to foregone earnings. For the workers in our sample, the present value of foregone earnings is Rs. 313,000. This implies a ratio of VSL to foregone earnings of 20 to 48-much higher than the range implied by Moore and Viscusi (1988) of 8 to 23 for theU.S. Two factors could explain this result. One is that Indian workers are more risk averse than U.S. workers. A number of studies examine consumption-saving decisions and the degree of risk aversion for the U.S. population. The values of the relative risk aversion index ( p ) that have emerged from this literature range between 1 and 7 (Laibson, 1998 and Heal, 1998). In contrast, very few studies examine the issue of risk aversion in developing countries. One study does provide some estimates of risk aversion for India, although in a much different context. Binswanger (1980) empirically estimates risk aversion among farmers in India from data on their investment decisions. His value of p, equal to 1.5, does not, however, adequately explain the discrepancy in the ratios. 1