What Are the Objects of Interest?

We have seen earlier that various approaches can be used to describe household behavior, from the unitary setting to noncooperative approaches and the collective model. Ultimately, the choice among these various frameworks will rely on particular considerations. First, general methodologic principles may favor one approach over the others. For instance, one can argue that the unitary framework is not totally faithful to methodologic individualism, a cornerstone of micro theory that postulates that individuals, not groups, are the ultimate decision makers. A second requirement is the model's ability to generate testable predictions for observable behavior that can be taken to data using standard techniques. Standard consumer theory fares pretty well in this respect. Utility maximization under a linear budget constraint yields strong predictions (e.g., adding-up, homogeneity, Slutsky symmetry, and negative semidefiniteness and income pooling), and adequate methodologies have been developed for testing these properties. Finally, a crucial criterion is the fruitfulness of the approach, particularly in terms of normative analysis and policy recommendations. A remarkable feature of standard consumer theory is that individual preferences can be uniquely recovered from demand functions (if these satisfy the Slutsky conditions); it is therefore possible to analyze welfare issues from the sole knowledge of observed behavior. This is a particular case of the general requirement that the model be identifiable, that is, that it should be possible to recover the underlying structure from observed behavior.

The first line of argument, concerning methodologic individualism, was evoked earlier. In this chapter we concentrate on the remaining two aspects, namely, testability and identifiability of preferences and processes from observed behavior. Most of the existing knowledge for nonunitary models concerns the cooperative framework and especially the collective model. The testability requirement per se is not problematic. The idea that a model should generate predictions that can be taken to data belongs to the foundations of economics (or any other science!).