3 Determining who is vulnerable

!UNDER CONSTRUCTION!

Estimating vulnerability using fused household surveys

Estimating the share of households vulnerable on any dimension requires “fusing” different data sources since not all dimensions are available from the same household survey. Data on income, education, access to water and access to electricity are mostly available from the same survey in GMD. Data on social protection, finan-cial inclusion, and non-income dimensions missing in GMD for a particular country are based on other surveys (Findex, DHS, MICS), censuses or administrative datasets. Dimensions of vulnerability considered from these datasets can be summarized at rural/urban, welfare quintile, and in some cases, subnational level. These disaggregated statistics are used to fuse information about the vulnerability of population subgroups into GMD surveys at household level.

The simple fusion approach proceeds as follows. For each dimension not in GMD, each household is randomly assigned as being vulnerable or not based on the rate of vulnerability observed in the strata it belongs to (as measured in the additional survey source). This preserves summary statistics for each population subgroup, such as, for example, the rate of account ownership among the poorest quintile reported by Findex. Second, the share of households vulnerable on at least one dimension, including existing and imputed variables, is calculated at the representative unit level (GSAP regions). This random assignment and aggregation process is repeated 100 times to account for household heterogeneity within each subgroup. The average share of households vulnerable on at least one dimension across all 100 simulations is used to calculate the indicator.

Estimating vulnerability using spatial data

The spatial data defining the accessibility vulnerability indicator is resampled (in the same manner as the expo-sure data) to align with the population grid. This classifies each rural grid cell as either more than 2 km or less than 2km distance to all-season roads. This is overlayed with the exposure data to determine the population exposed and vulnerable with respect to the spatial vulnerability dimensions.