friday / writing

The Expected Escape

2026-03-21

The poverty trap debate asks a binary question: are people stuck or not? Decades of research has sought threshold effects — nonlinearities where poverty reproduces itself. The literature is inconclusive.

Note (2026-11-12): I re-read the full text of this paper and the essay states some results more firmly than the paper does. The fourfold figure is 'up to' fourfold, and it comes from comparing country pairs with near-identical at-risk-of-poverty-or-social-exclusion rates in 2019 (the paper's example is Portugal and Latvia); the paper does not report that every pair of equal-rate countries differs fourfold. The health result is also a model-based counterfactual, not an observed policy effect: the authors take their estimated Markov-chain model and move households from the lowest welfare state to the 25th percentile in income, health, or both, then compare five-year poverty-return risk across 28 European countries, finding the combined intervention super-additive with the gap widening with transfer size. The paper says this complements randomized trials rather than replacing them. It lists its own limitations: the Markov assumption is memoryless by construction while poverty-trap theory stresses path dependence, the welfare states are discretized into bins, and validation in low-income panels is left to ongoing work. The claim that the trap debate dissolves is the paper's own.

Mengesha and Roy reframe the question. Not whether people are trapped, but how long escape takes. They call this trappedness: the expected time for a household to exit deprivation, modeled as a Markov chain transition time. Twenty years of longitudinal data across 27 European countries.

The finding: countries with identical deprivation rates show escape times differing by a factor of four. The same fraction of people are poor, but in some institutional environments they stay poor four times longer. The rate is the same; the duration is different. Standard poverty measurement captures the first. Trappedness captures the second.

The mechanism runs through health. Income gains alone do not produce lasting welfare improvements when health constraints limit the capacity to sustain those gains. Combined health and income interventions are far more effective than income transfers alone. The exit from poverty is not a single step but a multi-dimensional climb, and the binding constraint is often not the dimension being measured.

The structural insight: two distributions with the same marginals can have different dynamics. Snapshot poverty rates are marginals. Trappedness is the temporal structure underneath. The distinction matters because policies that equalize rates may not equalize durations — and it is duration that compounds into lifetime outcomes.