identical demand functions for each good in terms of p and z, as does the utility function F[ ψ (x, p, z), z] where F is monotone increasing in its first argument. Yet the two utility functions give different levels of utility; circumstances z can affect wellbeing without affecting observable market behavior—public goods whose costs are compensated could be an example—so that we cannot get from behavior to wellbeing without identification rules, essentially an exclusion restriction that z affects wellbeing only through its observable effects on behavior. It is not clear how such exclusion restrictions can be justified, and we know from Nicholson’s discussion of Engel’s own example that the exclusion cannot always be right. Direct observation of utility through happiness surveys would resolve the conundrum, provided that such observations do indeed correspond to standard concepts of utility, Daniel McFadden (2014). Household surveys continue to provide insights and to pose puzzles, particularly about food, which has been a consistent focus since the beginning. Here are some. There is a large body of evidence, primarily from differential mortality, that suggests that, in at least some of the countries of the world, girls are discriminated against in favor of boys. Amartya Sen’s (1989) documentation of “missing women” is the most famous. We might reasonably expect to see evidence of such discrimination in household surveys. The obvious place to look is expenditure on food, but surveys rarely attempt to find out who eats what. However, there is an indirect approach. Long ago, Erwin Rothbarth (1943) suggested that we might look at goods that were bought only by adults, tobacco and alcohol being the most obvious, and that expenditure on such items should indicate the expenditure needs of children. Given that children do not bring additional resources at birth, space needs to be made for their needs in the family budget, and so we might expect expenditure on adult goods to fall. In Deaton (1989b), I suggested that, in the presence of discrimination against girls, Rothbarth’s method should reveal that parents make more space in the budget for boys than for girls, so that, to take a concrete example, controlling for total expenditure, Indian parents of boys might spend less on their bidis than Indian parents of girls. Yet I found no difference, indeed relatively precisely so, a result that has been replicated in a number of settings, see Deaton (1997). It might well be that discrimination lies, not in food provision, but in the provision of medical care, for example—in countries with dowries, girls are long term liabilities and boys long term assets—but the method itself could (as usual) be challenged on various econometric grounds. Another food puzzle is related to economies of scale, the idea that two people can live more cheaply together than apart, so that members of larger households