To What Extent Do Poor Households Use Small, Routine Income Increases to Improve Their Diet?

When given additional income, how much do the poor choose to spend on food? This deceivingly simple question is surprisingly difficult to answer and has preoccupied many generations of economists. For starters, do observed correlations between income and food demand truly capture the effect of income on food demand, or do they also capture the effect of confounding factors?

In a new paper titled "Income and the Demand for Food Among the Poor," we revisit these age-old questions. To do so, we aggregate publicly-available data from five randomized evaluations of conditional cash transfers in five countries across three continents: two in Mexico, one in Nicaragua, one in the Philippines, and one in Uganda. We first define the demand for food as how much recipients spend on food (i.e., food expenditures). We then look at the impact of (i) being assigned to receiving a conditional cash transfer, and (ii) extra income on food expenditures overall as well as on expenditures on staples, protein, fruits and vegetables, and other foods.

Our pooled analysis (relying on both ordinary least squares and Bayesian hierarchical modeling)  shows that increases in income cause expenditures to increase across all food categories. Moreover, we conduct what believe to be the first credible test of Bennett's Law—which is the empirical regularity that poor households seem to respond to increases in income by (i) spending more on fine staples than they do relative to coarse staples, or (ii) spending more on protein relative to staples. However, empirical evidence for Bennett’s Law had hitherto consisted of correlations. In contrast, our setup allows us to perform a rigorous test of the law, showing that as income increases, consumers substitute fine grains for coarse grains and protein for coarse grains—but not necessarily for fine grains—providing partial support for Bennett's Law.

This is despite these cash transfers having two key design elements that makes them what is sometimes called “nutrition sensitive” by people who work in the field. First, all of them are targeted at women under the assumption that women will spend the cash on food and on investments in children while men will simply smoke or drink away the cash (note that the evidence does not bear out the latter claim!). Second, they all provide targeted information on optimal child nutrition to program recipients. If anything, that should mean that the households receiving these transfers are primed to spend the cash on nutrition, possibly making these elasticity estimates an upper bound of how food demand by the poor responds to income changes.

Even with “nutrition sensitive” programming and among some of the poorest people in the world—we find strikingly high levels of hunger in each setting considered—our estimated income elasticities for food are much smaller than many published estimates. But while our estimates leverage small and routine exogenous changes in income, previous estimates either rely on cross-sectional variation that is likely biased due to confounders or relies on a single but very large income shock. Our overall estimate of income elasticity of food overall is such that for a 1 percent increase in income, food expenditures increase by 0.03 percent. Our results thus suggest that while modest increases in income—whether through social transfers or economic growth—may help with reducing poverty, they may have a more limited impact on hunger or nutritional outcomes among the poor. These estimates provide fresh insight into the extent to which small income changes as provided by social safety nets and, possibly even through economic growth, may help progress towards the SDG 2 aim of zero hunger. They also inform the handwringing over the carbon impacts of increased demand from animal-sourced foods in low- and middle-income countries.