3.1 Drivers of Poverty: Context versus Structure Multidimensional poverty in Myanmar is shaped by the interaction of two distinct factors: structural household vulnerabilities and contextual conditions. Structural vulnerabilities refer to persistent socio-economic characteristics, encompassing educational attainment, employment type, income position and demographic composition, that shape a household’s underlying risk of deprivation. Contextual conditions refer to area-level factors beyond household control, including conflict intensity, geographic isolation, infrastructure quality, market access and service availability. In some areas, poverty is largely explained by these household characteristics. In others, poverty far exceeds what household profiles would predict, indicating that conflict, displacement, isolation or other area-level pressures elevate deprivation well beyond structural baselines. In a third group, households possess risk profiles typically associated with high poverty, but currently record lower deprivation levels, suggesting that favourable area-level conditions are moderating, though not eliminating, underlying vulnerability. To distinguish these patterns, observed MPI headcount was compared with predicted probabilities from a survey-weighted logistic regression model9 based solely on household characteristics. This comparison reveals three distinct poverty patterns:10 • • • Context-dominant poverty: Where observed poverty substantially exceeds predicted levels, area-level pressures, such as conflict, displacement, geographic isolation, infrastructure gaps or market disruption, push deprivation far beyond what household characteristics alone would produce. Structure-dominant poverty: Where observed and predicted poverty align closely, deprivation is primarily driven by entrenched household characteristics. Long-standing disadvantages, including low education, precarious employment, and limited income, explain the poverty pattern, rather than recent contextual shocks. High structural vulnerability (latent risk): Where predicted poverty exceeds observed levels, households carry high risk profiles (informal employment, low assets, limited education) but are partially protected by favourable area-level conditions, such as dense markets, service infrastructure, economic opportunities. These households are not currently classified as poor but remain vulnerable to deteriorating conditions. Union and urban–rural patterns At the national level, observed MPI headcount (32.6%) closely matches predicted probability (32.1%), indicating that structural household characteristics explain most aggregate deprivation. This national alignment, however, masks sharp divergence between urban and rural areas. 9 The predicted MPI headcount is estimated from a survey-weighted logistic regression model predicting MPI poverty status (binary outcome) from household characteristics alone: household head age, education level, employment type and occupation, household size, presence of school-age and pre-school children, and asset ownership. The model includes survey weights to account for PPS 2025 sampling design. For each household, survey-weighted predicted probabilities of being MPI-poor are calculated based on their characteristics. These predicted probabilities are then aggregated to the geographic unit (urban/rural, state/region) to produce predicted MPI headcount rates. The comparison between observed (directly measured from data) and predicted (model-based, holding area-level conditions constant) headcount isolates the structural component of poverty and makes explicit the contribution of contextual factors through the residual gap. 10 States/regions are classified based on the gap between observed and predicted MPI headcount (observed minus predicted), compared against a statistically informed threshold. The standard error of the predicted headcount is derived from the 95 percent confidence interval of the logistic regression model. States/regions are classified as context-dominant poverty if the gap exceeds +1.5 standard errors, as high structural vulnerability (latent risk) if the gap is less than −1.5 standard errors, and as structure-dominant poverty if the gap falls within the ±1.5 standard error range. L AYE R S O F DE P R I VAT I O N : M u l t i di m en si on a l Pov er t y i n M y anmar 19