THE THEORY AND CLASSIFICATION OF TARGETING

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 

A basic distinction in the targeting literature is between two forms of error,

that of undercoverage, that is the failure to reach some of the target group,

and of leakage, that is where benefi ts accrue to those outside the target

group. Following statistical terminology these are termed ‘type 1’ and ‘type

2’ errors, respectively. Practical application of targeting measures inevitably

involves some trade-off between these two errors. For example to minimize

undercoverage or type 1 error, more generous means of assessing eligibility

may be used, whilst to minimize leakage or type 2 error, stricter criteria

may be applied, and if these are not specifi ed or applied correctly they may

serve to exclude some of the target group. The social costs of the two types

of error need to be compared and arguably the poorer a society, the more

serious will be errors of omission or undercoverage relative to the costs of

leakage (Cornia and Stewart, 1993).

Figure 1.1 illustrates the alternative possibilities with the areas labeled

C and D corresponding to type 1 and type 2 errors, respectively. Another

way of expressing this information is to identify the targeting ratio, that is

the share of the non-poor (or non-target group) in benefi ts, relative to their

share in total population. The closer this ratio gets to unity, the weaker will

be the effectiveness of targeting.

In terms of theory, the comparison between a universalistic and a targeted

approach has been analyzed by Besley and Kanbur (1991) and here we

follow their presentation. If the poverty line is set at income level z and

individuals have incomes of y then an ideal targeting solution would be

transfer amounts of z–y varying between individuals depending upon their

initial income level. In this way all would be brought to the poverty line. The

costs of transfers would have to be borne by those above the poverty line.

Figure 1.2 illustrates this, showing initial income on the horizontal axis and

income after the transfer on the vertical. For points on the 45-degree line

Experiences with poverty targeting in Asia: an overview 3

initial and post-transfer income are equal. Line 1 shows the post-transfer

outcome in relation to initial income. Those below the poverty line z receive

a total transfer equal to the shaded area and those above face a tax, as their

post-transfer income is below their initial income. The fi scal cost will be

the sum of z – y for all in poverty initially. In contrast a universal approach

transfers the same sum to everyone. If poverty is to be alleviated fully and

the marginal poor person has zero income (because they rely on a share of

the family income) the transfer will be z per person and this will entail a

much higher fi scal cost, which will be z times the population.

Figure 1.3 illustrates this case with (as before) the shaded area giving the

amount of transfer. Those with initial incomes below z gain the full transfer,

whilst taxes are imposed on those above the poverty line, so their gain is z

minus the additional tax they must pay; at income level y* individuals will

Note: Targeting ratio is share of non-poor in benefi ts divided by share of non-poor in total

population, or B/(A+B) divided by (D + B)/(A + B + C +D).

Figure 1.1 Two types of error in poverty targeting

start to lose from the scheme as extra taxes will exceed z. There is leakage

to the non-poor at income levels between z and y*. Such leakage is the ‘type

2’ statistical error in the poverty literature, but undercoverage (or type 1

error) will be zero, as all are brought up to the poverty line.

Given the relatively high leakage and the fi scal costs involved, universal

transfers may appear obviously unattractive. However in any real world

situation there are also diffi culties with the ‘ideal solution’ of Figure 1.2.

• There are practical problems of lack of information concerning

beneficiaries, so that the initial incomes (the y’s) are not known

accurately. Hence the need for indicators of poverty that should be

correlated with income. We discuss below approximate ways used in the

past in the country cases to identify the poor. Where not all of the poor

can be identifi ed and reached, there are potentially serious problems

of omission from targeting schemes (the undercoverage rate), which

by defi nition should be absent in universalistic approaches. Hence, as

noted above, the social costs of errors 1 and 2 need to be compared.

• There can be costs to individuals of their participation in targeted

programs – for example psychic costs arising from social stigma or cost

in terms of time for travel or in the provision of information. In terms

Figure 1.2 ‘Perfect targeting’: the ideal solution

Experiences with poverty targeting in Asia: an overview 5

of the ideal solution in Figure 1.2 , if costs are c per person, then those

of the poor with an income above z–c will choose not to participate

in a targeted program, so that those with an income between z and

z–c will remain below the poverty line. Universal schemes may also

impose costs, and individuals may choose not to participate, but the

expectation is that such costs will be lower per dollar of benefi t.

• Incentive effects can undermine the impact of a fi nely targeted program

since in the ‘ideal solution’ the marginal tax rate is 100 per cent for the

poor. This arises since any shortfall in income below the poverty line

is to be covered by a transfer, and if incomes rise the transfer will fall

to match this. Hence if marginal tax rates infl uence the poor in their

productive activity there is a serious problem of dependence.

• Finely targeted schemes imply high administrative costs for their

operation and in general there will be an expectation that the more

fi nely targeted these are (that is the lower is the degree of leakage)

the higher will be the ratio of administrative costs to benefi ts to the

poor. This has the important implication that the optimal degree of

targeting need not be to aim for the minimum degree of leakage since

the costs of such targeting need to be compared with the benefi ts. We

can illustrate this simply in Figure 1.4 where the horizontal axis shows

Figure 1.3 A universal scheme

the degree of targeting – that is the share of benefi ts going to the poor

– from an intervention. This ranges from a low but positive fi gure

(T min), since the poor will gain something from any non-targeted

activity, to just below 100 per cent, since zero leakage to the non-poor

is implausible. The vertical axis gives monetary values of cost and

benefi ts per dollar received by the poor. Line A shows the relationship

between increasingly fi nely targeted interventions and administrative

costs per dollar of benefi t to the poor, which is assumed to rise steeply.

There will be some minimum cost required to establish any scheme,

which is shown as C min. Unit costs of targeting must be compared

with B, which is the marginal social benefi t of an extra dollar going to

the poor as compared with someone at the average income level. This

value must exist conceptually, provided benefi ts to the non-poor have

a positive social value (implying a trade-off between gains to different

groups). B is drawn as declining with the accuracy of targeting and

the intersection between the curves A and B gives the optimal degree

of targeting T*.

• Finally in political economy terms since only the poor gain, in

comparison with a universalistic approach there may be no infl uential

political constituency arguing for targeted schemes. This raises the

potential paradox that programs with a high leakage may have strong

Figure 1.4 Optimal targeting

political support, due to gains by a politically infl uential middle class,

which sustains a higher level of program expenditure than would be

otherwise possible. It is thus conceivable that in absolute terms the

poor may gain more from a universal scheme than from a more fi nely

targeted one.

The strong implication of these points is that whilst concerns over

leakage and budgetary costs may undermine the case for a universalistic

solution, methods of targeting must balance costs, associated particularly

with administration and incentives, against gains to the poor. The test for

targeting measures therefore becomes one of cost effectiveness where the

objective is to create income or income equivalent gains for as many of the

poor as possible at the minimum cost.

The country cases shed considerable light on issues of leakage and

undercoverage, but other aspects noted here remain unclear. For example,

there is little evidence from the country cases surveyed on the quantitative

importance of either costs to the poor from participation in targeting

schemes or of incentive effects. Also estimates of benefi ts to the poor, in

terms of income, consumption or welfare changes, relative to costs, are

rarely available to allow precise comparisons between alternative targeting

schemes. However in a few cases there are data on costs of transferring

income to the poor, for example from employment creation schemes or

food subsidies. In the absence of this type of data it is diffi cult to estimate

the optimal degree of targeting. Finally, the studies do confi rm the political

economy problem of generating support for targeting. The relatively low

amount of resources devoted to targeting schemes in all of the countries

indicates a problem with generating an infl uential political constituency

for these measures.