A cost-benefit framework for evaluating conditional cash-transfer programs

For analyses of these programs see: for Mexico, Schultz (2004); for Brazil, Bourguignon, Ferreira and Leite (2003); for Columbia, Attanasio, Gomez, Murgueitio, Heredia and Vera-Hernandez (2004); for Ecuador, Schady and Arajuo (2008); for Honduras, Morris, Flores, Olinto and Medina (2004); for Jamaica, Rawlings and Rubio (2005); and for Nicaragua, Maluccio and Flores (2004).

For CBA evaluation of non-CCT interventions for HIV/AIDS see Brent (2010a).

See Kakwani, Soares and Son (2005).

I. Dhaliwal et al. (unpublished manuscript) state at a number places in their paper that they are well aware of the need to use CBA rather than CEA. For example they write: “Even in the case of cash transfers it is not necessarily true that the marginal value of $1 to a poor household is equal to the marginal value of $1 to a wealthier household.” So they would seem to want to use unequal distributional weights in their evaluation. Yet they still recommend working within a CEA framework, where such weights play no role. They try to justify their non-use of CBA by pointing out that it is often difficult to value effects in monetary terms, which is true, but not a reason not to carry out a CBA. For some reason they think: (a) there is something wrong if one country values an outcome differently from some other country and so benefits are not standardized across countries, even though there is no reason why individual willingness to pay for outcomes would be the same, and (b) they want the preferences of the evaluator, and even the preferences of the reader of the evaluation, to count in the evaluation, see I. Dhaliwal et al. (unpublished manuscript), when neither of these preferences are usually a part of a social evaluation (where only those parties affected by the project itself are to count). They select CEA because the effect (a year of education) is the same in all education evaluations, and they intend in their paper to standardize the cost accounting, so the cost method would be the same in all countries. This would indeed achieve their purpose of making comparisons of education interventions across countries simple and transparent. But this comes at a great expense as it achieves the proverbial “throwing out of the baby with the bath water.” Nearly everything that demonstrates that one project is more socially worthwhile than another is excluded from the evaluation. Moreover the endeavor to just focus on costs per effect in a CEA of a CCT program leads to a host of complications. I. Dhaliwal et al. (unpublished manuscript) devote a lot of space examining whether the cash transfers themselves are costs or not, and so whether they should be included in the CEA calculation. Clearly the cash transfers are costs as someone has to give up resources to pay for the transfers. However, these transfers are also benefits (as our general CBA framework in Section 2 recognizes). Since there is no place for benefits in a CEA evaluation they are excluded in I. Dhaliwal et al.’s (unpublished manuscript) evaluation framework. This biases the results. For example, as we shall highlight in our discussion section, the Progresa CCT program gives out transfers that are five times as large as the Kenyan OVC CCT program, so the Progresa program will need to show effects that are over five times larger in order to display any cost-effective advantage over the Kenyan program. Another reason why their evaluation outcomes are biased is because they use a 10% discount rate, which is much too high. They use this figure because they base their determination of the social discount rate on the social opportunity cost rate, which is the wrong concept. As explained in Brent (2006, ch. 11), the correct concept is the social time preference rate.

The closest to a complete application of CBA to a cash transfer program is Coady’s (2000) evaluation of Progresa. The analysis is very comprehensive as it includes estimates of almost every ingredient of a CCT CBA including the size of the cash transfers, private costs, administrative costs, distributional weights and education rates of returns, and all of this is broken down according to targeting and conditionality so different CCT designs can be separately evaluated. The only problem with the study is that it shies away from doing a CBA at the very final stage and reverts back to doing a CEA. Coady suggests that a CBA would not be likely to give a different evaluation result as to the relative desirability of secondary school cash transfers over primary school cash transfers than is obtained from a CEA. But this can only be known for sure by applying the full general CBA framework presented in this paper. Multiplying the cost-effectiveness ratio by the value of the education effect (the net present value of the extra stream of income that comes from the additional year of education) to incorporate the benefits of education is an insufficient adjustment to covert a CEA into a CBA, see Coady (2000, p. 72), as it ignores the cash transfers as benefits and does not allow for distributional weighting, see our equation (8).

Considering the health conditionality in a CBA of a CCT does not add any new conceptual issues, for it is the sum of CBA ingredients for the two (health and education) combined that is to determine the desirability of the CCT. Focusing just on the educational component simplifies the exposition of this paper considerably. For methods to carry out CBAs of health expenditures see Brent (2003). Note that in the New York City CCT program analyzed by Riccio et al. (2010) there were as many as 22 different conditional incentives in the original version, including for the first time conditions related to the parents’ employment behavior.

In this criterion, the government cost of financing the program (the sum of the transfers and the administrative costs) has the weight a_NP. In this formulation we identify the general taxpayer as being the same as the non-poor group. Now it is true that the poor may pay some taxes and thus contribute to the government’s costs. But the non-poor generally pay the greater share. Moreover this is just a scaling issue. The weight given to the taxpayer, even though the taxes may consist of contributions by the poor as well as the non-poor, is given a weight of 1. This is the weight numeraire. It is the relative weight to the poor a_P that affects outcomes and this is generally >1.

Dividing w by the transfers will leave the CBA decision unaffected.

A reviewer has suggested that equation (8) be augmented by adding a_NP b^L to the benefits (because the non-poor may be attracted by the transfers to attend school and experience the long-term benefits) and –a_Pc^PR/t be added to the costs (because the non-poor may incur private costs). Note that these refinements are not likely to be quantitatively significant seeing that (as we shall see in our application) private costs and long-term benefits are likely to be small and the distributional weight to the non-poor is only going to equal 1. So we omitted these refinements from our general criterion. However, readers of this article are free to include these additional terms if they think that these refinements would affect the outcomes of a particular CCT program that they are considering.

In SSA there are often more poor persons than non-poor persons. So one should expect with random targeting, θ>0.5. This would then be a reason why targeting may not be so essential in CCT programs in this region.

Equations (9)–(12) give all the logical possibilities. In practice, not every ingredient would appear in every equation. For example, it can be expected that there will be no in (12) because there are no private costs with targeting per se. Thus in the comparative statics exercise in equation (13) this term does not appear. Similarly, with conditionality on its own the share going to the poor would not be expected to change. So θ₂= 0 and there is no θ term in equation (14).

Note that because we have formulated the CBA criterion per unit of transfer, the amount of the transfer is fixed per person. The differential dw does not have a dt term in it. So when we carry out our comparative statics with all the CCT design features below, although it is true that, say, changing conditionality would probably result in fewer recipients taking up the transfers, and so total cash transfers would fall, our criteria per unit of cash transfer would be unchanged. Effectively in our analysis we are changing the design features holding total transfers constant, i.e., dt=0. Therefore t is a parameter in all the expressions (9)–(12).

This way of applying the Squire and van der Tak formula is explained in Brent (2010b, p. 399).

See Bryant (2009, table 1).

Not surprisingly, as the main objective of CCT programs is to take recipients out of poverty, if one employed equal, unity weights, the net benefits of the overall Kenyan OVC program given by equation (8) would have been negative (equal to –0.79). The “switching value” (the value that just makes the evaluation negative) for the inequality aversion parameter applied to the overall evaluation is 0.57.

For a study that finds external health benefits of CCTs see Avitabile (2011).

Again, dividing w by 1+m makes no difference to the CBA criterion seeing that if w>0 then w/(1+m) will also be positive irrespective of the size of m (assuming that it is non-negative).

See Coady and Parker (2004).

For studies that have used non-unity distribution weights in the context of evaluating cash-transfers and have also employed the weighting formula specified in equation (15), and its variants, see Coady and Skoufias (2004) and Skoufias and Coady (2007).

For an early survey of schools on distribution weights, see Brent (1984). For further discussion and applications, see Brent (1998) and Brent (2006). For applications to health care evaluations, see Brent (2003) where distribution weights are examined in the context of CEA as well as for CBA. The most recent survey is in Brent (2010c).