Trappedness vs deprivation — MFPT uncouples from AROPE
Expected escape time (MFPT) and AROPE (at-risk-of-poverty-or-social-exclusion) are distinct: countries with nearly identical AROPE can have MFPT differing up to ~4×, implying static poverty rates miss persistence properties .
Example (Netherlands): MFPT from lowest welfare state = 3.16 years (pre-COVID) → 3.63 years (during COVID) indicating COVID deepened trappedness for low-welfare states .
Adding health to income lowers the estimated mixing time for the Netherlands (income-only mixing ≈ 42 years → income+health mixing ≈ 16 years), showing multi‑dimensional analysis can both reveal and unlock mobility pathways invisible in unidimensional views .
A common shock (COVID-19) altered not only household distributions but also the potential landscapes themselves, and did so differently across countries — evidence that trappedness is institutionally shaped rather than purely a household fixed trait .
Recovery times to the pre-pandemic steady state vary markedly across countries (roughly 2 years in best-case Scandinavian contexts to >10 years in parts of Southern/Eastern Europe), and do not simply track the magnitude of initial mobility shocks .
Mechanism: stronger social protection (unemployment insurance, universal healthcare, education funding) dampened basin deepening; fragile regimes exhibited bifurcations and new intermediate attractors that trap displaced households .
Health constrains returns to income — counterfactual interventions
Counterfactual: moving households from lowest welfare to 25th percentile in income, health, or both shows combined income+health interventions are frequently super-additive — the joint effect typically exceeds the sum of single-dimension effects across the 28 countries studied .
Health-only yields consistent five‑year risk reductions (~0.05–0.15) across countries; income-only effects are variable and sometimes near-zero (e.g., Denmark, Estonia), indicating income transfers alone may fail to produce durable exits when health is limiting .
Retention probability above median welfare increases with transfer size but is strongly stratified by initial health: at +5 percentile income boost, retention gap between healthiest and least healthy ≈ 5–7 percentage points; at +40 percentile it's >10 percentage points — health dependence rises with transfer size .
Pipeline: choose dimensions (income, health, education) → discretize (percentile or equidistant binning) → estimate transition matrix (MLE on longitudinal EU‑SILC microdata) → compute steady state π, potential Φ = −log π, MFPTs, mixing time, Shorrocks index, and simulate recovery by perturbing π with COVID-era transition matrices .
Key definitions: MFPT (M_{A→B}) measures expected years to exit poverty basin; mixing time quantifies institutional rigidity; recovery time t* computed by iterating pre‑COVID matrix on perturbed distribution until KL divergence falls below ε .
Robustness & limits: tests for curl (reversibility), memory length (k=1 vs k>1 marginal gains), bin-size sensitivity and bootstrapped uncertainty estimates are reported; data demands and Markov assumption caveats are discussed in detail .
极端数值与数据质量:某些国家在2D mixing time 或 MFPT 的估计呈极端值(例如 C Z 2D mixing ~459年),并伴随极宽置信区间(表 C6 标注若干国家置信区间异常宽),提示在这些案例中估计不稳、可能由样本量、分箱或记录错误导致,需在后续研究中用更多稳健检验或不同数据源验证。
結語(简短总结):
该报告通过把“被困”(expected escape time 等动力学量)作为与剥夺率并列的核心测度,构建了一套基于马尔可夫链与势能景观的定量框架,揭示了多维福祉交互、健康约束与制度架构在塑造贫困持续性中的关键作用,并为政策设计提供“先诊断拓扑、再设计干预”的结构性指导。该方法在欧洲高质量纵向数据中表现出显著诊断能力,但因数据、模型假设与因果识别边界仍需慎用与进一步验证。