Errors of Undercoverage and Leakage

К оглавлению
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 
17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 
34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 
51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 
68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 
85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 
102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 
119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 
136 137 138 139 140 141 142 

Aside from malpractice, which has been relatively common, if not always

well documented, in our country cases there are instances of what we can

term technical errors of targeting. This can be demonstrated most readily

for location targeting measures, since average income and consumption

estimates are normally available at the level of provincial or local government

units and these can be compared with national or provincial poverty lines

and with the allocation of public expenditure. Most studies indicate that

regional targeting has in practice been a relatively ‘blunt instrument’ for

reaching the poor.

For Thailand, we have detailed evidence from Warr and Sarntisart

(Chapter 5 in this volume), who examine the distribution of government

expenditure between rich and poor provinces, although they have no

information to allow an assessment of intra-province distribution. They

correlate provincial public expenditure per capita under different broad

categories with provincial per capita incomes, and fi nd positive elasticities,

so that in general expenditure per person and, by implication, benefi t rises

with income. Hence there is no evidence of progressive targeting across

provinces by broad expenditure category. When the same exercise is repeated

for the specifi cally poverty-focused expenditure no signifi cant relationship

with provincial income per capita is found for most categories. However,

provincial size does appear to matter so that, in general on a per capita

basis, smaller provinces are favored in poverty-targeted expenditure. Only

in the case of one minor category (the ‘Poor and Low-Income People’

expenditure) is there a signifi cant negative relationship between allocations

per capita and provincial income. This category was only 6 per cent of total

poverty expenditures over 2000–2002, and within it the clearest evidence of

a progressive allocation was for grants for health care. Hence on a regional

basis within Thailand, there is no evidence of a successful targeting at

poorer provinces.

For PRC, Park et al. (2002) and Wang (Chapter 4 in this volume) assess

what they term ‘targeting gap errors’ by examining the classifi cation of

counties as ‘poor’ in the light of their estimated income per capita relative

to the poverty line.13 What they term the ‘targeting count gap’ (TCG) can

be interpreted as the percentage of counties that are mis-targeted and this

can be disaggregated into the two types of error. Table 1.1 below shows the

situation taking the offi cial poverty line to estimate mis-targeting.

Table 1.1 PRC provinces: targeting count gap 1986 to 1995

Year Type 1 error Type 2 error Total

(undercoverage) (leakage)

1986 0.094 0.050 0.144

1987 0.082 0.065 0.146

1988 0.044 0.101 0.144

1989 0.056 0.096 0.152

1990 0.078 0.093 0.171

1991 0.058 0.101 0.158

1992 0.038 0.107 0.145

1993 0.002 0.225 0.227

1994 0.005 0.232 0.237

1995 0.004 0.218 0.222

Source: Park et al. (2002), Table 4.

The data have an intuitively clear interpretation showing that the

effectiveness of targeting has decreased over time. Initially undercoverage

was the major problem, but over time leakage became considerably more

important, particularly after the re-designation of poor county status in

1993, when about 20 per cent of counties with incomes above the poverty

line became mis-targeted. However even with perfect designation at the

county level there would still be targeting errors due to the presence of the

non-poor in poor counties and of the poor in non-poor counties. Estimates

suggest that the share of the poor (at the offi cial poverty line) living in non-

poor counties rose from 29 per cent in 1992 to 38 per cent in 2001 (Wang,

Chapter 4 in this volume).14

Further evidence of errors in regional targeting comes from the

Philippines. Balisacan et al. (2000) identify the 25 most depressed provinces

in the late 1990s ranked both by the incidence of poverty or by the poverty

gap measure (the rankings are not identical). These are then compared

with the priority provinces under the Social Reform Agenda of the Ramos

administration. Out of the 26 priority provinces only 11 are in the ranking

of most depressed by the poverty indicators. It is clear that formal poverty

data were only one of a number of factors used by the government to

determine priority status.

Similar recent assessments of regional targeting for Indonesia are

unavailable, however survey work illustrates the error of omission in the

national Neglected Village program 1994–96 (IDT). As noted above, this

was a location-targeting program designed to channel small-scale credit

to the poorest households targeted at over 20 000 ‘neglected’ (that is poor)

villages across the country. Using a pilot study of the IDT in 384 villages

in 6 provinces Sumarto et al. (1997) demonstrate the weakness of targeting.

They illustrate undercoverage by focusing on the provinces of Central Java

and West Nusa Tenggara (WNT). In the former, 30 per cent of all villages

are classed as neglected and covered by the program, but 46 per cent of

the poor (insofar as these can be identifi ed accurately) are in villages that

are not covered. In WNT a much higher proportion of all villages are

classed as neglected, but still over 40 per cent of the poor live in non-IDT

villages and are not covered by the program.15 In addition for Indonesia,

the National Economic Survey (SUSENAS) provides detailed information,

which has been used to assess who has benefi ted from the set of povertytargeting

measures introduced in the wake of the Financial Crisis (Perdana

and Maxwell, Chapter 3 in this volume). Table 1.2 summarizes the results

of the most detailed study based on this data.

The data are extremely detailed and reveal clearly that of the anti-poverty

programs over the period only the subsidized ration scheme reached a

signifi cant proportion of those eligible (40 per cent). Subsidized rice reached

over 50 per cent of households in the bottom quintile, but for all other

schemes the proportion of the target group reached was below 20 per cent

and often well below it. Hence undercoverage was clearly a problem. In

terms of leakage this was most serious for the rice and nutrition programs,

where gains to the richest 20 per cent were high and the ratio of non-poor

benefi ciaries to their share in total population was highest (nearly 1.0 for the

nutrition program implying nearly zero targeting effectiveness), although

these fi gures do not reveal the magnitude of gains per family, only whether

they were in receipt of some benefi ts.

Table 1.2 Indonesia: Impact of anti-poverty programs August 1998–February 1999

Program Potential Coverage Coverage Coverage all Proportion of Targeting

recipients Poorest Richest potential benefi ciaries ratioa

(million) 20% (%) 20% (%) recipients (%) from non-poor (%)

Subsidized riceb 50.4 52.6 24.3 40.1 0.74 0.92

Employment creationb 50.4 8.3 2.5 5.6 0.70 0.88

Primary scholarshipsc 29.7 5.8 2.0 4.0 0.71 0.89



scholarshipsc 10.4 12.2 4.9 8.4 0.71 0.89



scholarshipsc 6.4 5.4 2.0 3.7 0.71 0.90

Health cardsd 27.6 10.6 3.1 6.3 0.67 0.83

Nutritione 20.0 16.5 14.2 15.9 0.79 0.99


a Targeting ratio is share of non-poor (defi ned as those above bottom quintile) in total benefi ciaries to their share in total population, which is 0.80 by

defi nition.

b Subsidized rice and employment creation programs potentially available to all households.

c Scholarships are potentially available to all individual pupils enrolled at the relevant levels.

d Health cards potentially available to all those individuals who were estimated to have visited a health care provider in the three months prior to the


e Nutrition support potentially available to all individuals in the ‘pregnant women and children under three years’ category.

Source: Sumarto et al. (2001)


Self-targeting schemes were intended to overcome many of the problems

faced by directed or narrow targeting. Nonetheless they have also proved

disappointing in many cases. In India there has been considerable experience

with food-for-work and employment creation programs designed to attract

the poor by offering below market-clearing wage rates. Evaluations have

revealed serious undercoverage. In the 1990s the Employment Assurance

Scheme offered on average only 17 days of employment per person per

year against a target of 100 days. Further, its village coverage was low

with another evaluation fi nding no more than one third of eligible villages

actually covered. This meant that in some states less than 10 per cent of

the target group was reached. This, combined with the low number of days’

work on offer under the scheme, rendered its overall impact on the welfare

of the poor largely minimal. In this case part of the problem had to do with

the slow release of central government funds to the states and part to lack

of matching funding by the states themselves (Srivastava, Chapter 2 in this

volume). In other schemes, however, the level of wages set for employment

has been identifi ed as a critical factor with relatively high and therefore

attractive wages leading to a ‘crowding out’ of the poor. In India under

the food-for-work scheme in a survey in Andhra Pradesh, Deshingkar and

Johnson (2003) conclude that wages either in cash or in kind were set too

low in prosperous villages thus attracting non-poor migrants, but too high in

poorer villages leading to crowding out of the poor. A similar conclusion is

reached for an Indonesian employment creation scheme (the Padat Karya).

An evaluation of this, drawing again on the SUSENAS data, found that

for the 1998–99 period, as many as 70 per cent of benefi ciaries were from

non-poor households (Perdana and Maxwell, Chapter 3 in this volume).

Self-targeting has also been implied by health and nutrition and many

micro-credit schemes. For example, in Indonesia the poor are entitled to

health cards giving them access to free medical treatment. The defi nition

of the poor was based on the BKKBN classifi cation scheme noted above.

Insofar as the better off will prefer to pay for improved access to health care

there is an element of self-targeting in such a measure. Initial assessments

of the Health Card program in its fi rst six months of operation, again using

SUSENAS data, showed substantial undercoverage with only around 10

per cent of the poorest 20 per cent of households covered. A subsequent

more detailed analysis suggested that even though coverage may have been

low, the scheme did help to prevent a decline in use of health facilities by

the poor in the wake of the Financial Crisis (Pradhan et al., 2002). More

explicit self-targeting is involved in the Affordable Medicine for All (GMA)

program in the Philippines which provides free drugs for a limited number of

conditions at public hospitals and a limited number of distribution outlets,

to which it is expected only the poor will choose to go for the drugs. There

is no fi rm evidence on the undercoverage or leakage associated with this

scheme (Balisacan and Edillon, Chapter 6 in this volume).

Micro-credit programs aimed at the poor have a substantial element of

self-targeting insofar as they involve the potential embarrassment of clients

being associated with poverty programs and the inconvenience of frequent

group meetings. Micro-credit is seen by many in the development community

as an important innovation in the fi ght against poverty (Morduch, 2000).

There is now considerable evidence that micro-credit has had a positive

impact on poverty reduction in a number of countries, although often it is

not the ‘core poor’ who are the main recipients, but rather those close to

or just above the poverty line. In terms of our case-study countries (and

elsewhere) there is also evidence of some leakage from micro-credit programs

(Weiss et al., Chapter 7 in this volume). However, this leakage appears

to be much less than from conventional subsidized credit programs. For

example, for PRC the subsidized loan program available for poor counties

went principally to economic entities rather than poor households (although

formally it was an obligation that recipient enterprises should have at least

50 per cent of their employees who were below the poverty line). Many

of these loans went to Township and Village Enterprises in poor counties

and the direct link with poverty reduction came to be questioned. The

introduction of micro-credit schemes in PRC in 1997 was a direct response

to this concern (Wang, Chapter 4 in this volume). In the Philippines an

assessment of the main low interest credit program for the poor (the Tulong

sa Tao program) of the Aquino administration concluded that targeting

was vague and that only around one-third of benefi ciaries were from lowincome

groups (let alone being amongst the core poor) (Balisacan et al.,

2000). Similar assessments are given for such schemes in India. For example,

an assessment of the Integrated Rural Development program, which was

designed to provide subsidized credit to the poor for income-generating

activities, found that in the states of Bihar and Jharkand, 24 per cent of

benefi ciaries were above the poverty line and a high proportion had incomes

just below it (MAKER, 2003).

Finally, broad targeting based on types of expenditure that the poor

will use disproportionately offers an alternative to the type of narrow

targeted schemes discussed above. Assessing the impact of measures like

health and education expenditure is normally done by ‘benefi t incidence

analysis’ (van de Walle, 1998a). A typical conclusion is that primary health

care and primary education expenditure have a disproportionate positive

effect on the poor, whilst expenditure on hospitals and higher education

have a disproportionate positive effect on the better-off.16 The net effect

of aggregate health and education spending will vary therefore depending

on how expenditure is allocated within the sector, but in general there is

evidence that broad targeting within these sectors can reach the poor.

In terms of other evidence on the impact of broad categories of investment,

a simulation exercise for the Philippines, using coeffi cients derived from a

regression model of poverty, shows general road expenditures to have high

economic returns, but to have a negative direct effect on the poor, although

this is compensated by a positive impact from growth. Electrifi cation emerges

as the best option in terms of high economic returns and a relatively strong

positive effect in reducing poverty (Balisacan and Edillon, Chapter 6 in

this volume).