What Poverty Metrics Miss When We Treat Them as Interchangeable

Most studies of poverty and child development need a single variable for poverty, so they pick one. An asset index because the survey had one. Consumption because the economists asked for it. A multidimensional index because it feels more complete. The choice is usually made early, defended briefly, and then quietly carried through every result that follows.

The problem is that these measures do not agree with each other. Households that look poor on assets are not always the households that look poor on consumption, on housing quality, on food security, or on access to services. When you line up the rankings side by side, the overlap is real but partial. Which means the measure you chose is doing quiet work in your findings.

In my dissertation work in rural Côte d'Ivoire, I spent a lot of time on exactly this question: which dimensions of household poverty actually track children's learning and wellbeing, and what happens when they are collapsed into one number. The honest answer is that collapsing hides variation that policy audiences care about. A programme that improves household assets is not the same as a programme that improves food security, even when both move a composite index by a similar amount.

There is a practical version of this argument and a conceptual one. The practical version is that reporting sensitivity across measures should be as routine as reporting standard errors. If a result only holds under one poverty definition, the reader deserves to know. The conceptual version is harder: our measures encode assumptions about what deprivation is, and those assumptions travel badly between a village in Côte d'Ivoire, a neighbourhood in Mumbai and a district in Philadelphia.

None of this is an argument for measurement nihilism. Indices are useful precisely because they simplify. It is an argument for treating the choice as a research decision with consequences, documenting it, testing it, and being willing to say that the answer depends on how you ask the question.