Skip to main content

DHS Geospatial Covariates (GC)

Product status: CLUSTER-LEVEL COVARIATE SILVER IMPLEMENTED
Authority: L3 REBUILT / SOURCE-DERIVED CLUSTER MEASUREMENT
Real protected-data acceptance: NOT RECORDED IN THIS SITE
Experiment-use status: NOT YET A CURRENT DHS EXPERIMENT INPUT
Producing repository: fcv-empirical-data

What it is​

DHS GC is represented as covariate measurements associated with DHS survey clusters.

It is not represented as a generic polygon-by-period covariate panel for Africa.

The durable path is:

external DHS GC release file(s)
        ↓
SourceSnapshotRef + survey/file linkage
        ↓
wide source-native cluster Silver
        ↓
optional derived long Silver
        ↓
variable temporal semantics + availability QA

Natural grain​

The authoritative wide Silver table is:

survey_id × cluster_id

A derived long view may use:

survey_id × cluster_id × source_variable

The long view is derived from the hashed wide Silver product. It is not a second fictional Bronze source.

Source and cluster identity​

The current implementation preserves DHS cluster identifiers as identifiers rather than allowing them to become covariate measurements.

In particular, DHS identity fields such as DHSID and DHSCLUST remain available for survey/cluster linkage but are excluded from the covariate variable set. Explicit attempts to select identity fields as measurements are rejected.

The source release, snapshot identity, reversible source-native columns, and wide → long lineage remain durable.

Geography​

GC values are attached to DHS clusters. The product does not claim that those values are authoritative polygon-wide measurements.

No GID aggregation is performed by the GC vertical.

A later scientific use may link a GC cluster to analytical geography through the separate DHS GE/GPS product, but the GC value remains a cluster-associated measurement unless a named downstream derivation says otherwise.

Temporal semantics​

GC temporal meaning is documentation/registry-driven rather than inferred from convenient filenames or survey year.

The shared survey temporal vocabulary includes:

  • static;
  • survey_time;
  • annual;
  • epoch;
  • climatology;
  • retrospective;
  • unknown.

No rule match means unknown. Conflicting rules remain unknown and are reported.

A year is assigned only when a documented rule explicitly licenses parsing or supplies a documented year. A variable without a year suffix does not automatically receive the survey year. Even survey_time does not require fabricating a measurement_year when survey timing already exists separately in the catalog.

Implausible parsed years remain visible as parsing/QA states rather than being repaired by substituting survey year.

Missingness and coverage​

Coverage is cluster measurement availability, not polygon or country coverage.

The current semantics make two distinctions explicit:

  • a missing cluster-variable row is not observed;
  • a present cluster-variable with missing source value is a missing source measurement.

Neither is zero.

The GC adapter does not fill, interpolate, forward-fill, back-fill, or replicate static values across periods.

What this product deliberately does not do​

The GC vertical does not:

  • reconstruct source rasters;
  • run zonal statistics;
  • build an Africa-wide background covariate panel;
  • aggregate to GID;
  • turn static values into annual series;
  • assign survey year implicitly;
  • construct _DHSGC compatibility panels;
  • choose regression covariates;
  • assign treatment/outcome/control roles.

Experiment history​

No current contract-backed DHS experiment has yet consumed these GC products.

A downstream design must still decide which GC variables, if any, serve as controls, moderators, descriptive features, or other scientific roles. Their source temporal semantics constrain that use but do not determine it.

Technical reference​