PUBLIC WORKS AS AN ANTI-POVERTY PROGRAM: AN OVERVIEW OF CROSS-COUNTRY EXPERIENCE by K. Subbarao Paper submitted for the invited session of ASSA Annual Meetings, January 3-5, 1997. K. Subbarao is a Principal Economist in the Poverty Analysis and Social Assistance Group, Poverty and Social Policy Department of the World Bank.
Public Works As An Anti-Poverty Program: An Overview of Cross-Country Experience Introduction. Public works program has been an important counter-cyclical intervention in developed and developing countries. The program was used in some western countries during the depression years of the 1930s. More recently, several developing countries in Asia, Africa and Latin America have adopted the program in some form or another. In terms of person days of employment created, the programs in Bangladesh and India are perhaps the largest in the world. Employment in India s nationwide program (known as Jawahar Rojgar Yojna, or JRY) reached a billion workdays by 1995. This paper begins with a brief discussion of the role of public works as a safety net and then presents the available cross-country experience on various aspects of the program. The conclusions are then summarized. Public Works: An Overview of Conceptual Issues. The argument for public works in 19th century England centered around the ethic of work--often long and nasty work. The program is playing a different role in low-income developing countries. Its role is that of a safety net -- conferring transfer and/or stabilization benefits to the poor, while at the same time using the poor s labor to build infrastructure for development. The program s effectiveness depends on the benefits (transfer as well as stabilization benefits), costs, and the way resources are raised to finance the program. What guidance can theory provide us on these aspects?
The transfer benefit to a worker from a day s employment in public works amounts to the wage he gets from the scheme, net of any costs incurred both the costs of participation (such as the cost of transport) and the earnings lost from alternative employment (Ravallion, 1987; Datt and Ravallion, 1992). If the costs of participation and income from alternative sources are negligible, and the program has no effect on the labor market and the structure of wages, the transfer benefit approximates to the program wage times the duration of employment. Quite plausibly, the assumptions are unlikely to hold in reality. Unless the scale of the program is very limited, the program is likely to put an upward pressure on the wage rate, thereby reducing the demand for labor. Moreover, the costs of participation and foregone earnings are only rarely zero. Depending upon the impact of the scheme on the wage rate, the foregone earnings, and the costs of participation, the transfer benefit may be higher or lower than the program wage. 1 Apart from the wage rate, the share of wages in total cost, and the duration of employment determine the total transfer benefits going to workers. One argument advanced is to keep the program wage low so that the poor are selfselected into the program. A low wage rate is most likely to keep the nonpoor out of the program. However, a low wage will also result in lower transfer earnings per (poor) participant. It keeps the overall participation rate low, while at the same time ensuring a disproportionate number of poor participants, than would be the case if the wage rate were higher. Given a budget, a relatively high wage rate is most likely to lead to rationing of jobs under the scheme, apart from attracting nonpoor workers to the program. The potential income gains from a works program can be exaggerated if the source
of financing is ignored. If employment is expanded entirely through an aid-financed public works program, the advantage (transfer benefits) to workers is straight-forward. However, if the program is financed out of general tax revenues, it would be important to examine: (a) the counter-factual situation, viz., would the participants of the program have benefited more (less) from alternative ways of spending the same budgetary resources, and (b) if a works program has expanded at the expense of other activities conferring nonlabor income for the poor (i.e., educational or hospital services). Rarely does one find empirical evidence on either issue. Yet, a discussion of how a works program is financed is important for an understanding of the true benefits of the program. The stabilization benefit of the program depends on its timing. For example, if the program s timing synchronizes with the agricultural slack seasons when the demand for labor is low, workers are most likely to gain from the resulting income stabilization and consumption-smoothing. Stabilization benefits from unanticipated declines in farm activity in bad agricultural years may be even higher than from anticipated declines in slack seasons. Even if the transfer benefits are small, income stabilization can pre-empt acute distress and prevent poor households from onerous adjustments such as distress-selling of productive assets in bad agricultural years. In other words, the risk benefits conferred by a works program via its insurance function can be as important as the transfer benefits to the poor via its distribution function. Finally, the particular design features, the institutional framework and the type of implementing agencies (e.g., line Ministries of the government, private contractors, nongovernmental agencies or a combination of the above) also influence how much benefits the program is in fact transferring to the poor. If the implementing agencies and the
institutional framework are strongly attuned to leakages, the poor who participate in the program may have to pay a part of wages either to officials or contractors or politicians, or incur other transaction costs that may reduce the transfer benefits of the program. But the design of the program can also be rendered beneficial for certain vulnerable groups. For example, the mode of payment of wages cash or in-kind (food) is an important design feature that influences participation decisions. Payment of wages in-kind or piecewage payments may attract more women than men to worksites. The program features (especially the level of the wage rate and the timing of the program) and the design features (implementing agencies and the institutional framework) together determine the program s efficacy as an anti-poverty intervention and its costeffectiveness. The cost-effectiveness of the program can be gleaned from the cost per job created, and the cost per one dollar of transfer to the poor. The next section discusses cross-country experience with respect to financing wage rate and targeting, duration and timing of employment, share of wages in total cost, cost per job created, and the impact on poverty and incomes. Public Works: Cross-Country Experience. (a) Financing and scale of operations. Sources of financing influenced both the duration and character of programs. Thus in the 1950s and 1960s, bilateral food aid (mostly wheat under US PL 480) financed food-for-works programs in India and Bangladesh. As food aid diminished, food-forworks was replaced by cash for work. The program began to be funded entirely from domestic sources in India, accounting for its sustainability. By contrast, in most countries of Africa, bilateral donors funded many public works programs. Donor funding meant a
net (additional) transfer of income to the poor. But the programs were shortlived, lasting only as long as donor support lasted. Depending upon historical circumstances and source of financing, scale of operations varied across countries (Table 2). (b) Wage rate and targeting. The wage rate is a key element determining the distributional outcomes of the program. One important way to ensure that the program reaches the poor is to maintain the program wage at a level no higher than the ruling market wage rate for unskilled labor. This practice also helps keep overall costs down. In most countries, governments set the levels for minimum wage for unskilled labor. Though minimum wages are legally binding, their enforcement varies, especially within large countries. Whether the minimum wage is higher or lower than the ruling market wage and whether or not it is enforced, are empirical questions; even in the same country the relationship between the two might vary over time and, in large countries, across regions. In countries where the market wage is less than the minimum wage, the poor may be selfselected to the program if the program wage is set at a level no higher than the ruling market wage, even if it were lower than the statutory minimum wage. However, political and legal constraints may make it difficult to maintain the program wage at levels less than the minimum wage. Not surprisingly, there is much cross-country variation (Table 1). The wage rate was set at 70 percent of the minimum wage in Chile; it was about 25 percent higher than agricultural market wages in the Philippines; it was equal to the statutory minimum wage in Kenya. In India, the nationwide program of JRY maintained its wages at the statutory minimum wage, though market wages were much lower than the minimum wage in many regions. In the Maharashtra Employment Guarantee Scheme
(MEGS) also, the program wage was equal to the minimum wage. But the minimum wage remained lower than the ruling market wage until 1988. In 1988, the minimum wage as well as the program wage were doubled. This upward revision of the wage rate in 1988 contributed to job rationing, thereby eroding the "guarantee" of employment expected of the program (Datt and Ravallion, 1994). In Tanzania and Botswana, because the program wage rate was maintained at a level higher than it was in comparable unskilled activities, jobs had to be rationed particularly during droughts when the poor's need for participation in public works was the greatest (Teklu, 1994b). The program wage in Kenya and the Philippines was also found to be higher than the market wage for unskilled labor; as a result substantial numbers of the nonpoor were attracted to the program (Teklu, 1994a, Subbarao, et. al., 1995). But in some countries (Burkina Faso, Senegal and Sri Lanka), the program wages were lower than market wages. The lesson: a relatively high wage rate is most likely to result in job rationing. (c) Duration and timing of employment. While a lower program wage is most likely to render the program self-targeted, its careful timing (so as to synchronize it with the agricultural slack season) might also help in enhancing stabilization benefits. India's JRY, though generating over 800-850 million person-days of employment annually, conferred relatively small transfer benefits inasmuch as the person days of employment per person as well as per household remained modest often no more than five per family per month (Dev, 1996). This small level of job creation is clearly insufficient to raise the incomes of poor families. Nevertheless, the program operated intensively during the agricultural off-peak season: as many as 55 million persons gained employment in the agricultural off-peak seasons. Because of the timing of the program, "consumption-
smoothing" (stabilization) benefits were considerable, despite its other disadvantages. In MEGS also, Walker and Ryan (1990) show that the risk benefits conferred by the scheme are considerable. Although jobs were rationed and the total number of person days of employment sharply declined after the wage rate doubled in 1988, stabilization benefits of the scheme continued as the MEGS operated intensively in off-seasons in both 1980-81 and 1990-91 see Figure 1. In Kenya and Tanzania, not only were program wage rates higher but the timing of public works was synchronized with the busy agricultural season, thereby significantly diminishing both the transfer and stabilization benefits to the poor (Teklu, 1994a, 1994b). In Bangladesh, too, the food-for-work program was poorly synchronized with seasonal patterns of need for work and food. (d) Share of wage cost in total cost. In most programs, the share of wages to total cost of programs varied between 0.3 to 0.6 (Table 2). While a higher share is desirable for higher transfer benefits to be conferred on the poor, many factors determine this ratio including the nature of the asset to be created, the duration and timing of the works and, most of all, the availability of technically and economically feasible labor-based methods of production. In most road construction activities, the share of wages to total cost ranged between 0.4 to 0.5 percent. Where the work has been entrusted to private contractors, the outcome with respect to the share of wage cost to total cost is unpredictable. Evidence from Ghana suggests that timely availability of project funds is important for the adoption of labor-based methods (Stock and de Veen, 1996). (e) Cost per job created. Various factors influence the cost per job created including the mix of locals and expatriates involved in the implementation of subprojects; the delivery mechanism selected; particularly the modalities of hiring private contractors;
the wage rate; the capital-intensity of operations; and administrative capacity. Disaggregated information on the above factors is hard to come by; so it is difficult to disentangle the various factors underlying the variations in cost per job created. Available data on cost per job created, which belong to late 1980s or early 1990s, are shown in Table 2. There is much cross-country variation -- from as low as $1 per person day of employment created in Bangladesh to $8 in Bolivia. To place cross-country comparisons in perspective, mean consumption per person per month are also shown. The main finding is that not all countries managed to create jobs at low cost. Low-income developing countries in Asia and Africa Bangladesh, India, Botswana, Tanzania have incurred a cost person day in the (modest) range of $1 $2. Honduras and Chile in Latin America have incurred a lower cost per job created than other countries in the region. The variations in the cost of job created seem to widen when evaluated in PPP exchange rates. Table 3 shows data on annual cost per job created in public works, evaluated at PPP exchange rates in seven countries. The cost appears to be much too high for Senegal and Ghana. Few studies have examined the cost per dollar of income transferred through public works programs. The data released by Concurrent Evaluation Report, Ministry of Rural Development, suggests that India s nationwide program of JRY transferred one rupee of income to participants at a cost of Rs 1.9 (including the amount of transfer) in 1991. It is well-known that JRY participants included the poor as well as the nonpoor, largely because the program wage was higher than the market wage. Taking into account the extent of mistargeting, the program required an expenditure of Rs 4.35 (including the amount of transfer) to transfer one rupee of income to a poor participant.
(f) Impact on incomes, poverty and social life. Evaluations of works programs in general, and Maharashtra Employment Guarantee Scheme in particular, suggest significant transfer and stabilization benefits. Numerous micro studies have shown that 60 to 70 percent of the participants of India s nationwide program (JRY) and the MEGS belonged to poor households. The immediate net income gain from works programs as a percent of total gross earnings is high. For example, the severity of poverty has fallen from 5.0 percent to 3.2 percent owing to participation in the EGS (Datt and Ravallion, 1992). Dev (1992) notes significant social gains: The EGS also discourages sexual barriers and inequality... Women now dress better and their economic power has given them a better status in their families. (p. 54). Datt-Ravallion (1992) show that workers in EGS did forego earnings from alternative sources, but the foregone earnings are not substantial. Nevertheless, their findings suggest that the opportunity cost of a participant s time needs to be factored in while designing the program. Foregone income can be lowered by more flexible timing, by adoption of piece-rates and task-based mode of payment, providing work closest to homes, and by expanding the program in agricultural slack seasons. Women s participation seem to be particularly sensitive to the flexibility in timing and location of work. IV. Conclusions Three general conclusions can be drawn from the reviewed experience. First, the level of the wage rate is critical in determining the distribution of benefits from the program, as well as how much of the program is targeted toward the poor. Maintaining the program wage at the level no higher than the ruling market wage for unskilled labor can enable the poor to self-select themselves into the program.
Second, the timing and duration of employment determine the nature of benefits offered by the program: in general, the more narrowly the program is restricted to operate only during agricultural slack seasons, the lower the transfer benefits and the higher the stabilization benefits. Even if transfer benefits are small, the program s stabilization (risk) benefits can be substantial. However, a public works program entirely for consumptionsmoothing purposes is more appropriate where poverty is transient the numbers of the vulnerable rising during bad agricultural years and in slack seasons and falling in good
years and peak agricultural seasons than in countries with chronic and severe poverty. In countries subject to serious interyear fluctuations (e.g., Ethiopia), and in countries where poverty gap ratios are high, the need for rendering the program year-round (thus raising the transfer benefits to the poor) assumes greater importance. Third, the program can be so designed as to encourage greater participation of women, a higher degree of involvement of the private sector and lower transaction costs for the poor to participate in the program. 11-A
References Datt, G. and M. Ravallion, 1994. Transfer Benefits from Public Works Employment: Evidence for Rural India. Economic Journal 104: 1346-69. Datt, G. and M. Ravallion, 1992. Behavioral Responses to Workforce: Evidence for Rural India. LSMS Working Paper, World Bank, Washington, D.C. Dev, S. Mahendra. 1996. India s MEGS: Lessons from Long Experience. Indira Gandhi Institute of Development Research. Bombay. Mimeo. Ravallion, M. 1987. Market Responses to Anti-Hunger Policies: Effect on Wages, Prices and Employment. WIDER Working Paper Number 28. Stock, E. and J. de Veen, 1996. Expanding Labor-Based Methods in Road Programs. World Bank, Africa Technical Department. Washington, D.C. Mimeo. Subbarao, K., A. Bonnerjee, J. Braithwaite, S. Carvalho, K. Ezemenari, C. Graham, A. Thompson. 1996. Social Assistance and Poverty-targeted Programs: Lessons from Cross-Country Experience. World Bank, Poverty and Social Policy Department. Washington, D.C. Draft. Subbarao, K., A. Ahmed and T. Teklu. 1995. Philippines: Social Safety Net Programs: Targeting, Cost Effectiveness and Options for Reform, Discussion Paper Number 317. World Bank, Washington, D.C. Teklu, T. 1994a. Minor Road Program in Kenya: Potential for Short-term Poverty Reduction through Asset Creation. International Food Policy Research Institute. Washington, D.C. Mimeo. 11-B
Teklu, 1994b. Labor-Intensive Rural Roads in Kenya, Tanzania and Botswana: Some Evidence on Design and Practice. International Food Policy Research Institute. Washington, D.C. Mimeo. 12-A
Footnotes K. Subbarao, Principal Economist, The World Bank. 1 For an analytical model that clarifies the issue further, see Ravallion (1987). 12-B
Table 1: Public Works: Program Wage (PW), Minimum Wage (MNW) and Market Wage (MW) in Selected Countries Country/Program PW in Relation to MNW and/or MW 1. Bangladesh: Cash For Work, 1991-92 PW<MW 2. India: (a) Cash For Work, JRY, 1991-92 (b 1 ) MEGS: up to 1988 (b 2 ) After 1988 PW=MNW>MW PW=MNW<MW PW=MNW>MW 3. Pakistan: IGPRA* III, 1992 PW<MW 4. Philippines: Cash For Work 1990 Food For Work 1987 PW>MW PW**>MW 5. Botswana: Cash For Work PW<MNW, but >MW 6. Kenya: Cash For Work, 1992-93 PW=MNW>MW 7. Chile: Cash For Work 1987 PW<MNW=MW Source: Compiled from Subbarao, et. al.. (1996). * IGPRA: Income Generation Program For Refugee Areas, funded by UN Agencies. ** Program wage includes cash wages plus the monetary value of food supplied by the program. 12-C
Table 2: Public Works: Scale of Operations and Costs for Selected Countries Scale of Total Cost (wage & Operations nonwage) per person Ratio of Mean Consumption Program/Country/Year (million person day of employment Wage Cost to per person per days annual) created Total Cost month (year) US $ % US $ 1. Bangladesh: 1991-92, FFW 15 1.6 0.5 46.85 (1990) 2. India: 1991-92, CFW (JRY) 850 1.3 0.6 28.40 (1990) 1991-92, MEGS 100-180 1.2 0.51 3. Pakistan: 1992, CFW (IGPRA) 5.15 2.8 0.6 4. Philippines: 1990, CFW 0.3 3.2 0.5 61 (1988) 5. Botswana: 1992-93, CFW 7 1.7 0.63 6. Ghana: 1988-91, CFW 0.5 3.4 0.2 73 (1988) 7. Kenya: 1992-93, CFW 0.6 3.0 0.3-0.4 58.6 (1992) 8. Bolivia: 1982-90 CFW 8-9 8.0 0.4 86.2 (1990) 9. Chile: 1987, CFW 40-45 0.5 131.4 (1990) 10. Honduras: 1990-91, CFW 2.5 1.0 0.4 42.8 (1990) 11. Costa Rica: 1991-94, CFW 8.9 4.0 99.7 (1990) Note: FFW: Food For Work; CFW: Cash For Work JRY: Jawahar Rojgar Yojna, a nationwide program of public works MEGS: Maharashtra Employment Guarantee Scheme IGPRA: Income Generation Program for Refugee Areas, funded by UN agencies. Source: Compiled from K. Subbarao, et. al. (1996). 13 13
Table 3: Annual Cost of Job Creation in Public Works under Different Social Funds in US$/Job Egypt Honduras Nicaragua Madagascar Bolivia Senegal Ghana Cost per Job at Market 1,401 2,120 2,580 786 2,700 5,445 2,122 Exchange Rates Cost per Job at PPP 7,212 9,759 14,302 3,620 9,388 12,100 10,610 Exchange Rates Source: Bank staff estimates, as cited in K. Subbarao, et. al. (1996). 14-A
Figure 1. Labor Attendance under the EGS 16 1980/81 1981/82 1982/83 1983/84 1984/85 1985/86 1986/87 1987/88 1988/89 1989/90 1990/91 8,000 14 Total Labor Attendance 7,000 12 6,000 Percentage 10 8 6 4 Percent distribution, 1980/81 5,000 4,000 3,000 2,000 Person days in thousands 2 Percent distribution, 1990/91 1,000 0 April May June July August September October November December January February March 0 14-B 13