EDUCATIONAL CASH TRANSFERS AND INTRA-HOUSEHOLD TIME ALLOCATION 7
members as single individuals bargaining over resources. They seek to endo-genize bargaining power in the household using either co-operative (McElroy and Horney, 1981) or non-cooperative game solution concepts (Lundberg and Pollack, 1997). Empirical tests comparing unitary and collective models have focused on two restrictions deriving from the unitary model. First, unitary models are based on the assumption that only total household income af-fects household demand. On the other hand, in collective models individuai income will affect the bargaining power of household members and, therefore, resource allocation. Another prediction of unitary models is the equality of cross-substitution effects: an increase in the income of each one of the spouse should have the same effect on the other spouse's labour supply.3 As highlighted in the preceding section, Oportunidades pays the transfer to the mother of the beneficiary child but we also observe a number of cases where the beneficiary is a male. Given that the transfer is likely to generate a shift in power distribution in the household, it may seem possible to use the collective household model framework to test for the income pooling hy-pothesis. Furthermore, since Oportunidades transfer is an exogenous source of unearned income variation, this looks like the best possible scenario to test the collective model. Indeed, the literature usually tests the prediction of the unitary model by comparing the effect of income in the hand of each spouse on the demand of some consumption good using either earned income4 or some form of unearned income.5 It has been argued that unearned income is a more credible basis for income pooling test than earned one since it eliminates the potential endogeneity problem.6 More recendy, some authors were able to exploit naturai or randomized experiment to find exogenous sources of decision power variation so that no possibility of correlation between household beha-viour and income is left.7 However, the variation in beneficiary recipient in our sample cannot be used to test income pooling because the recipient of the cash transfer is not randomly chosen. Even though the transfer may actu-ally increase women's bargaining power due to the higher share of income in their control, we cannot distinguish between variation in the outcome of in-terests due to this effect from variation due to different households prefer-
3	For an example of a study rejecting the equality of cross-substitution effects see Ashenfelter and Heckman (1974).
4	See Phipps and Burton (1992) and Bourguignon et al. (1992).
5	See Thomas (1990) and Schultz (1990).
6	Earned income is endogenous with respect to the resource allocation process in the household: the number of hours worked will be determined simultaneously with the consumption choices that the pooling test examines.
7	For naturai experiments, see Lundberg and Pollar (1997).