Assessment of the impact of targeting measures

PRC adopted a multi-level targeting mechanism, in which regional poverty targeting plays a fundamental role in the allocation of poverty reduction funds. In this section, I first examine the effectiveness of the regional poverty targeting through the analysis of the accuracy of poor county designation and the equity of poverty fund distribution, and then discuss the impact of targeting measures at project level.

Accuracy of the Poor County Designation

Targeting Gaps

Initial evidence on targeting can be found in the frequency distributions of poor and nonpoor counties across income levels. In 1986, only half of the counties in the lowest income decile were designated as poor, even though there were even more counties designated as poor in the next income group (Figure 2). In 1993, many fewer counties in the lowest income groups were being excluded--better coverage, but there were many more counties designated as poor in the middle-income groups--greater leakage (Figure 2).

To evaluate overall targeting effectiveness, Park, Wang and Wu (2002) defined new measures, which were referred to as targeting gaps and targeting errors. Targeting gaps describe mistargeting in the full sample with respect to a reference poverty line, while targeting error describes mistargeting given a set number of targeted beneficiaries. Similar to poverty measures, gaps and errors can be aggregated using different weights.

Two types of targeting gaps were calculated: the targeting count gap (TCGt) and the targeting income gap (TIGt). The targeting count gap is defined as

Here, N is the total sample of counties, indexed by i. I_it1 is an indicator variable for type I error (or incompleteness) that equals one if a county is not designated as poor (P_it=0) but its income per capita (Y_it) is below the poverty line (Z_t). Iit2 is an indicator variable for type II error (or leakage) that equals one if a county is designated as poor (NP_it=1) but its income per capita is above the poverty line. TCG_t can be interpreted as the percentage of counties that are mistargeted, and is easily disaggregated into type I and type II error.

The targeting income gap is defined as

Where the indicator variables are as defined above. It is similar to the TCG except that mistargeting is weighted by the magnitude of mistargeting, measured as the difference between income and the poverty line. The TCG and TIG are analogous to the widely used poverty headcount and poverty gap measures, but are two-sided rather than one-sided.¹²

Yearly TCG and TIG measures for PRC's poor county designation are presented in Tables 11 and 12. Both measures are sensitive to the chosen poverty line; as the line is increased type I error increases and type II error decreases. The authors calculated the TCG and TIG for each year from 1986-1995 for two different lines--the official poverty line and a relative poverty line equal to 60 percent of mean income per capita.¹³

The results show that targeting effectiveness has deteriorated steadily over time, that incompleteness has fallen while leakage has increased, and that using the official poverty line, targeting gaps jumped noticeably after the new poor county designations in 1993. As seen in Table 11, the percentage of counties that were mis-targeted increased from 14 to 22 percent using the official poverty line and from 15 to 19 percent using the relative poverty line. While failure to designate a poor county as poor was nearly twice as likely as designating a nonpoor county as poor in 1986 (using either the official or relative poverty lines), by 1995 the opposite was true using the relative line and virtually all mis-targeting was due to leakage using the official line. Considering that about one fifth of counties are mis-targeted, the TIG of 77 yuan in 1995 for the official line implies that the average magnitude of mis-targeting in mis-targeted counties is about 385 yuan, or nearly two thirds of the poverty line¹⁴. Only part of the targeting gaps can be explained by preferential treatment towards minority and revolutionary base counties. In 1986, 25 percent of leakage (type II error) in the TCG (using the official poverty line) was due to minority counties and 35 percent to revolutionary base counties. By 1995, the comparable figures were 35 and 19 percent.

Table 11[PDF: 375kb] | 60 pages]

One problem with the targeting gap measures is that they are sensitive to the number of poor counties designated. If the number of designations is less than the number of truly poor counties, type I error is unavoidable, and if designations exceed the number of poor counties, type II error is unavoidable, even when targeting is perfect in that designations go to the poorest counties. Another way to assess targeting, then, is to compare outcomes with the perfect targeting case given the number of poor county designations. The authors defined targeting count error (TCE) as the percentage of designations not given to counties that would be targeted under this definition of perfect targeting, or

Here, Z_t* is the income level of the marginal, or threshold, county when targeting is perfect given the number of available designations (D). Similar to targeting gaps, the indicator functions can be weighted by income differences with counties that were mistakenly targeted to calculate targeting income error (TIE_t) or by rank differences to calculate targeting rank error (TRE_t).¹⁵ These statistics (and formal definitions) are reported in Table 13, and show that by any measure, targeting error was substantial in the original designations (in fact, a majority of designations were mistargeted), increased steadily over time, fell dramatically after new designations in 1993 to levels even below that of the original designations, and then began increasing once again. Thus, the 1993 designations reduced targeting error, but through a strategy of expanded coverage beneficial to counties above the absolute or relative poverty thresholds.

Table 12[PDF: 375kb] | 60 pages]

Table 13
National Targeting Error, 1986-1995

Even if poor county designation was perfect, there would still be mistargeting due to:

The non-poor in poor counties
To the extent that poverty programs are targeted at poor counties, there is a danger of leakage to the nonpoor who live in designated poor counties. Total rural population in 592 counties is around 200 million, while poor population is only 28 million estamated by the official poverty line or less than 100 million mesured by $1 dallor/day stardanrd. Even if all of PRC's poor were resident in poor counties, the majority of households in poor counties would still not be poor.

The Poor in Non-Poor Counties
Data from the NBS indicate that, of the 80 million rural poor in 1992, only 23 million lived in non-poor counties, accounting for 29%. However, proportion is increasing. Estimate from the same source suggest that the poor living in non-poor counties account for 38% of total poor population in 2001. Rural household data provide evidence that a much larger percentage of the poor live in non-poor counties. A study indicated that about half of the poor in four southern provinces do not live in poor counties (Ravallion and Jalan, 1999).

Empirical Analysis of the Determinants of Poor County Designation

As we have known from the discussion of the poor county selection process above, status as a minority or revolutionary base county will have a significant effect on poor county designation. In 1990, 637 counties in PRC were minority counties (33 percent) and 195 were revolutionary base areas (10 percent). 20 percent of minority counties and 44 percent of revolutionary base counties were designated as poor in 1986, accounting for 38 and 30 percent of all poor counties. In 1993, the number of minority counties designated as poor more than doubled (to 46 percent of all minority counties) but the number of revolutionary base counties increased only slightly (to 48 percent). As a share of all poor counties, however, the number of minority and revolutionary base counties fell to 30 and 16 percent in 1993 because the total number of poor counties increased by so much.

Using county-level economic data from the Ministry of Agriculture, which were the basis of poor county designations in 1986,¹⁶ Park, Wang and Wu (2202) studied the determination of poor county status by estimating probit functions for poor county designations in 1986 and 1993. The results shed light on the first two targeting criticisms only. Explanatory variables include log of income per capita, log of grain production per capita, and industrial share of total income in the year preceding the designations, status as a minority county or revolutionary base county, and provincial dummy variables. All explanatory variables have estimated coefficients that are statistically significant. The fitted probabilities correctly predict the status of 92 percent of county designations in 1986 and 88 percent in 1993.

The marginal effects on the probability of poor county designation at the sample means for poor counties are presented in Table 14. In 1986, a 1 percent increase in income per capita reduces the probability of being designated a poor county by 1.3 percent, a 1 percent increase in grain output per capita decreases the probability by 0.2 percent, and an increase in the industrial share of income of 1 percent reduces the probability by 0.7 percent. Designations are less responsive to per capita income and grain production in 1993 (1.1 and 0.1 percent) and slightly more responsive to industrial share of income (0.8 percent). Being a minority or revolutionary base county increases the probability of designation by 15 and 45 percent in 1986, and 17 and 18 percent in 1993. Overall, the responsiveness of poor county designation to both economic and political variables decreases in 1993, mainly because the larger number of designations reduced the sensitivity of designations at poor county means. A comparison of elasticities evaluated at full sample means finds greater responsiveness in the latter period.

Table 14
Marginal Effects on Probability of Poor County Designation
(From Probits Evaluated at Poor County Means)

Many provincial dummies have large and significant coefficients, suggesting that there was considerable discrimination against specific provinces. In the 1986 designations, poor provinces in Sichuan, Guizhou, Yunnan (southwest), Henan, Hunan (central), and Inner Mongolia, Gansu (northwest) were at a severe disadvantage, while a county was much more likely to be designated as poor if it were in the wealthier provinces of Fujian, Shandong, Hubei, or Xinjiang. The starkest contrast is between Gansu and Fujian: a county in Gansu was 70 percent less likely to be designated as poor than a county in Fujian. In 1993, despite a large number of newly designated counties in relatively disadvantaged provinces such as Yunnan and Guizhou, southwest provinces remained at a distinct disadvantage, along with Qinghai and Ningxia in the northwest and Anhui and Hunan in central PRC.¹⁷ Many favored provinces in 1986 no longer appeared favored in 1993.

The Chinese experience confirms the view that regional targeting may be a rather "blunt instrument" for reaching the poor (Ravallion and Lipton, 1995). Even when funds are perfectly targeted at the poorest regions, there is considerable leakage to the nonpoor in poor regions and lack of coverage of poor in nonpoor regions. In PRC, political factors have strongly influenced poverty targeting from the outset.

Entrenchment of political interests to maintain poor county status has made adjustments difficult, leading to a tendency to expand coverage and increase leakage. Combined with the finding by Ravallion (1993) that Indonesia’s pattern of regional disbursements is poorly targeted, the evidence presented here suggests that political constraints are likely to undermine regionally targeted programs when the level of targeting is at the county level or higher.

Equity of Poverty Fund distribution

After poor county designation, poverty funds from different sources are delivered to those designated poor county through various channels following different regulations. The central government usually allocate the majority of poverty funds to the provincial government, and the latter then allocate the funds both from the central government and from its own budget as matching funds to the county government. Criterions adopted by the provincial government in deciding the amount allocated to each poor county varies considerably between provinces. In this section, I will discuss the determinants of poverty fund allocation among poor counties after a brief description of the main fund sources.

Sources of poverty reduction funds

The OLGPR categorizes three kinds of funds as rural poverty reduction funds in PRC, i.e. subsidized loan, fund for food/cash for work and budgetary fund. Total amount spent by the central government every year since 1986 is presented in table 15. Total nominal poverty funds increase steadily over the past 17 years, from RMB 4.2 billion yuan to RMB 29.1 billion yuan, or increasing at an annual rate of 12.9%. But the funds increase much more slowly in real terms, only from RMB 4.2 billion to RMB11.4 billion yuan, or increasing at an annual rate of 6.4%. Because of high inflation in late 1980s and early 1990s, fund amount in real terms stagnated till 1996. Only after 1996, had poverty reduction funds increased dramatically. Among the three funds, subsidized loan accounts for 59% of the total, FFW fund takes the second position with a share of 24% and the budgetary fund only account for 17%. Compared with the central government budget and GDP, poverty investments' shares are 5% and 0.2% respectively over 1986 and 2002 period. The shares of government budget are relatively higher in early and late 1990s and the shares of GDP are highest in middle 1980s¹⁸.

Table 15[PDF: 375kb] | 60 pages]

In addition to these three funds, several other sources are also important for poor counties in PRC. The fund for Compulsory Education Project in Western Regions from the central government totaled RMB 8.9 billion yaun. A recent study reveals that poverty investments from local governments and government departments equal one fourth of the investments from the central government (Li Zhou, 2001). International organizations such as the World Bank, UNDP, IFAD and bilateral development agencies such as Ausaid, JICA, DFID also have different kinds of poverty reduction projects in PRC for years¹⁹.

Distribution of poverty funds among poor counties

From a simple plot of average funding levels for the three programs during 1994-96 against income per capita, it is obvious that there is not a strong relationship between funding levels and income per capita (Figure 2).²⁰ The nonparametric estimate reveals a weak inverse relationship.

Using the NBS county level data collected from the Rural Poverty Monitoring Survey and OLS regression model, I test the extent to which county funding amounts from different sources for the period 1998-2001 can be explained by county characteristics for the sample of poor counties where data exists²¹. Provincial dummies are included in all regressions. Regression results are presented in table 16.

Table 16 Determinants of poverty reduction fund allocation (1998-2001)

The estimation results suggest that, except investment from other sources, fund allocations are significantly and positively related to the level of poverty incidence. One percent point increase in poverty incidence will increase the total poverty investment per capita by RMB 0.76 yuan, or total investment, subsidized loan, budgetary fund and FFW fund from the central government by 0.85, 0.46, 0.18, 0.21 yuan respectively. Total rural population has significantly negative impact on the allocation of all poverty funds, indicating large counties are at a disadvantage. Although revolutionary base counties are favored in poor county designation, they are discriminated against in fund allocation, as revolutionary base counties receive 26 yuan per capita less than non-revolutionary counties. Minority counties are still at a advantage, receiving 14 yuan per capita more than non-minority counties. Inland border counties are also favored in poverty fund allocation. Compared with counties with plain geography, funds allocated to counties in highland and mountainous areas are not significantly different.

Many provincial dummies have large and significant coefficients, indicating some provinces are at an advantage while others are discriminated against. Compared with Hebei province, Shanxi, Inner Mongolia, Liaoning, Fujian and Yunnan are at a disadvantage, while Jilin, Heilongjiang, Anhui, Jiangxi, Henan, Hubei, Hunan, Guangxi, Hainan, Chongqing, Sichuan, Gansu, Qinghai, Ningxia and Xingjiang are all in favor of, e.g. Qinghai receive 211 yuan per capita more from the central government even after controlling for poverty incidence, population size and minority status.

Effectiveness of community and household targeting

Due to data availability, few empirical works have been done on the effectiveness of poverty targeting at the community and household levels. I mainly resort to our field interviews with local officials, households and other anecdotal evidence for the discussion of targeting effectiveness at the project level.

Subsidized loan

PRC's subsidized loan scheme was widely criticized for failing to target the poor effectively. To a large extent, the problem was due to the political and economic environment in which Chinese local government institutions operate.

One broad source of targeting problems stems from the dual goals of the program—to reach the poor and to promote economic development. Serving two masters can lead to conflicts. First, to provide incentives for effective loan use and repayment, local poverty officials often use past performance as a criterion competing with poverty status in awarding new loans to lower levels. Second, many local officials believe that the poor are incapable of managing projects successfully and prefer to promote economic development by lending to enterprises, economic entities and large farmers.

Even more important sources of poor targeting arise because of factors motivating local officials. There are three local players with a stake in the use of subsidized loans: the local OLGPR, the local government, and the Agricultural Bank. Local poverty officials may compromise targeting objectives to meet the dual goals of the program. Local government officials also are concerned with generating revenues and furthering overall economic development, not just in poor areas and poor households, which may lead them to support diversion of funds to enterprises or investment in more promising regions. This is especially true given the acute fiscal woes of local governments in poor areas. Agricultural Bank officials are interested in profit and so care about loan repayment above all else. As the transaction cost of small loans to poor households was relatively high and loan use was difficult to supervise, neither the Agricultural Bank nor the Agricultural Development Bank²² was willing to grant loans to poor farmers in the absence of stringent supervising mechanisms. Because they disburse the funds, they can veto projects proposed by the local Poverty Alleviation Office if they feel the likelihood of repayment is low. This has led to numerous conflicts between bank officials and poverty officials. Even when loans are approved, Agricultural Bank officials have an incentive to shorten the period of the loan (so that funds can be relent quickly at higher rates), delay loan disbursement, or divert loans outright. Finally, rent-seeking and corruption sometimes led to diversion of the inexpensive subsidized loans to influential and ready-to-bribe non-poor groups. The record of subsidized credit programs in other countries also shows that in one way or another, influential interests rather than the poor often end up benefiting most from subsidized loans. There is much evidence attesting to the widespread leakage of subsidized loans.

The diversion of subsidized loan to non-poverty reduction activities has become more serious in recent years with the commercialization of ABC and the shutdown of most ABC township branches. A recent survey by the MOF find that the majority of subsidized loan are made to large scale enterprises and for infrastructure construction such as highway. In 2002, of 750 million yuan subsidized loan made in Jiangxi province, only 150 million were households (not necessarily the poor households) loan. Pingjiang county in Hunan province and Suichuan and Le An county in Jiangxi province have not made any loan to poor households in recent years (Wen, 2003).

Even when loans are lent directly to households in poor villages, in many cases they are not given to the poorest households. Evidence from a nationwide survey of villages conducted in 1996 provides some evidence on the targeting effectiveness of subsidized loans within villages. Of the 184 villages in 6 provinces that were surveyed, 32 had received poverty loans a total of 58 times in the past. Of these 58 times, data on the average wealth of households exist in 33 cases. Village leaders were asked whether most loans went to better-off farmers, average farmers, or poorer farmers. 57.5 percent of the time the loans went mostly to farmers of average wealth while 42.5 percent of the time they went to farmers of below average wealth. In no cases did village leaders report that the loans went mainly to better-off farmers. Also, the relative frequency of giving loans to average rather than poor farmers appears to have increased in the 1990s. Loans received before 1990 went to poor households 45 percent of the time. Loans received in 1990 and after went to such households only 36 percent of the time (Rozelle et al, 1999).

Food for work

One reason FFW programs have been praised and targeted for expansion is that because the funds bypass local budget bureaus, to date relatively few funds have been diverted for other uses, which has become common for many earmarked budgetary items, especially in poor counties (Park et al., 1996). There is concern that expanding the scope of FFW will make it more difficult to monitor and increase the incentives for local governments to think of ways to divert the funds to other uses, decreasing the programs effectiveness. Another disadvantage of creating too many channels for funding local agencies is that it hurts the transparency of the fiscal system, and effectively reduces the planning authority and capacity of local units.

Another targeting issue related to fund diversion is the possibility that local governments are substituting FFW projects for other funds that would have gone toward infrastructure construction. In other words, is it the case that FFW funds merely displace other funds and so do not greatly increase the amount of infrastructure constructed? There is little evidence to quantify the extent of such crowding out, but given the acute fiscal pressures hitting poor areas, some crowding out is likely.

A number of targeting issues also surround the actual FFW projects. These include the location of the projects, whether laborers are paid, and who participates in the construction work. Because FFW projects are investment projects, as with subsidized loans, local leaders inevitably balance the economic return of projects with the effect of projects on helping the poor. The return to building a road to a very remote village, for example, will be extremely low given the sparse populations served and the high cost of constructing roads in mountainous terrain.

Provincial poverty officials reported that in addition to poverty status, other criteria used in allocating FFW funds included the quality of project design, the ability of local leaders, and past performance. In some provinces, such as Henan, before 1994 some projects were awarded to nonpoor counties (though often with poor townships), but since then all funds have been allocated to national or provincial poor counties. Some county officials, however report that amounts awarded to different counties depend more on project feasibility and quality than on poverty status. This is likely to be even more true within counties. Zhu and Jiang (1995) report that villages that have greater population, favorable environmental conditions, more surplus labor, and which are more remotely located are more likely to be involved in projects.

One important issue in assessing the poverty alleviation role of FFW projects is the cost borne by local residents in the form of uncompensated labor effort. Because funds are limited, in many areas FFW funds are used to pay for material supplies while labor is supplied through yiwugong (essentially a labor tax). In some areas with FFW projects, the amount of yiwugong may surpass regulated limits (usually a maximum of 30 labor days per year). Even when workers are paid, the amount is often lower than the going wage. These costs to the poor in the form of foregone leisure or other income-earning activities must be weighed in assessing how well the programs are targeted and how much they benefit the poor. It is entirely possible that the labor tax associated with FFW projects taxes the poor more, since they are more likely to have surplus labor that can be tapped for construction work.

Zhu and Jiang (1995) report that 40 percent of households in their sample (in Sichuan, Ningxia, and Shandong) worked without receiving any pay. Older, male laborers with less land and more education are more likely to participate in projects. They also found that for most laborers (78 percent), time spent working on FFW projects did not detract of income-earning activities but rather decreased leisureconsumption only.

In other countries, especially India, public employment schemes offer work to anyone willing to work at the stated wage, which is purposely set fairly low. In this way, self-targeting is achieved because only the poor are willing to work at such a low wage. Voluntary participation also ensures that participants are better off from participating in the project, independent of the benefits from infrastructure construction. In PRC, however, because labor participation often is not voluntary and frequently uncompensated, participation is often a tax which must be weighed against the benefits from infrastructure. Were PRC to attempt to more fully incorporate self-targeting and voluntary participation into the design of its FFW program, the amount of infrastructure constructed would likely be reduced, but targeting might be improved.

Budgetary development fund

Of the three main poverty programs, least is known about the distribution and use of budgetary development funds because of the classified nature of budgetary data in PRC. Still, a number of targeting concerns warrant mention. First, because the development fund program began before the designation of poor counties in 1986, many of these funds were and continue to be given to counties not officially designated as poor, which may increase coverage but also increase leakage. Second, just as for Food-for-Work funds, it is likely that poor counties will substitute development funds for other budgetary resources that would have been allocated for similar purposes, reducing the impact of such funds on realized investment. Assuming perfect fungibility, development funds at worst act as a pure budget subsidy, so should help local governments in poor counties meet their own fiscal agendas, even if these lack the development focus that central leaders would prefer. However, if these transfers also affect the subsistence transfers negotiated between levels of government, the crowding out problem could be much more severe. In poor areas where budgetary crises have forced governments to delay or suspend wage and otherpayments to cadres, there also is the danger that funds will be diverted to non-productive use altogether.

Regulations stipulating that development funds be used to benefit poor households by developing projects probably prevent full crowding out as described above. However, local governments have a much stronger influence on the use of these fund than subsidized loans or Food-for-Work funds, so the danger of biases toward revenue-producing enterprise investments is greater, even though the success rate of such projects and their benefits to the poor are much less. Another concern is that when development fund is used in the areas of education for the construction of schools, it is usually required that villages to collect supplementary fees from the households to finance the financial gaps. This will have negative impact on poverty reduction in the short term.

Download this Discussion Paper [ PDF 375KB| 60 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.