Wealth-anchored category validation: A diagnostic for categorical orderings in fragile-state household surveys

Asset-based wealth indices and binary "improved" or "clean" service indicators both assume that the response options of a categorical survey variable are coded in the wealth order they imply. The assumption is rarely tested. I set out a procedure, Wealth-Anchored Category Validation (WACV), that tests it, and I am careful about what a failed test means. WACV ranks a variable's categories by their average position on a wealth index built from the other indicators, so no variable is judged against itself, and attaches stratified cluster-bootstrap intervals in which the whole estimation chain, including the index itself, is re-estimated in every replication. It reports two things: whether the categories sit on the local wealth gradient in the order the coding assumes, and whether a binary someone has already built from the variable falls where that gradient actually steps. Divergence on either is a diagnostic signal with several possible explanations: a coding fault, an analytical recoding fault, a context-dependent classification, or noise. The procedure does not adjudicate between them from wealth data alone. Applied to five ordinal variables in three waves of the South Sudan High Frequency Survey, the ordering diagnostic finds no robust inversion in the main specification; equivalence tests and a simulation locate what that null can and cannot rule out. The cut-point diagnostic finds one severe anomaly: an improved-housing binary used in earlier analysis of these data places the two poorest dwelling types on the improved side, and sits 0.26 standard deviations below the households it excludes. That anomaly replicates in a survey wave not used to develop the method. A codebook audit shows the binary originates in analysis rather than in the survey instrument, which makes it an instructive case of analytical recoding rather than a defect in the data.