LoginSign Up
← Back to all sparks
D

dscore

INFRA · APIS
Velocity0.0

D-Score for Child Development

github.com/d-score/dscore ↗

The D-score reference implementation rebuilt its measurement foundation on seven countries.

child-developmentpsychometricsglobal-healthr-packagemeasurementbreaking-changes
◆Current state
dscore computes the D-score and DAZ, the GSED developmental measurement used in child-health research, and it is the reference implementation rather than one option among several. The package is at 2.1.0 after a dense 2025: the default key moved from three-country to seven-country validation data, the licence moved from AGPL to Apache 2.0, and the 2.1.0 line added per-country references, full BSID-III coverage and domain-level scoring. Breaking changes are routine here and always come with a documented fallback key or algorithm argument.
◆Where it's heading
The arc runs from correcting the instrument to broadening who can use it. The 2020-2024 releases were item-table repair and error correction, including a scale-factor bug that altered published standard errors; from 1.11.0 onward the work is distribution — a permissive licence, more instruments mapped in, and references resolved per country rather than pooled. That combination points at national-survey and app-embedded use rather than research-only use.
◆Prediction
The 2.0.0 notes state that groundwork was laid for extending D-scores to older children, and 2.0.0 still tells users to fall back to gsed2406 for instruments outside GSED SF and LF. Expect the next releases to close that gap by mapping more instruments into gsed2510, with the older-age extension the likeliest headline feature.

◆Recent moves

  1. 7mo ago

    dscore 2.1.0

    The seven-country data that 2.0.0 made the default is now selectable as seven individual site references, so DAZ can be computed against a national rather than pooled distribution. BSID-III support is completed with 242 new itembank entries, ddomain() adds domain-level scoring, and the who_descriptive reference is renamed with a breaking-change note. Substantial work, but it extends the foundation 2.0.0 laid rather than moving it.

    View source ↗
  2. 10mo ago

    dscore 2.0.0

    ⚡ SPARK

    The release the rest of the current line depends on: the default key becomes gsed2510, built on the full seven-country GSED validation study instead of three countries. Everything in 2.1.0 — per-country references, wider instrument mapping — is this foundation being exposed.

    View source ↗
  3. 10mo ago

    dscore 1.11.0

    ⚡ SPARK

    The licence change that precedes the 2.x expansion: AGPL to Apache 2.0, made to meet Gates Foundation open-access policy. The same release more than doubles Bayley-III item matching, from 67 items to 172, raising by3 D-score estimates by roughly 2.6 D.

    View source ↗
  4. 1y ago

    dscore 1.10.0

    Prior mean and standard deviation become properly configurable, with a dedicated vignette, and the defaults for missing ages change from 50/20 to NULL so that absent ages fail visibly instead of being silently imputed. The preliminary_standards reference is refreshed on a larger Bangladesh sample.

    View source ↗
  5. 2y ago

    dscore 1.9.0

    A frank correction: a long-standing scale-factor error had made item characteristic curves look steeper than they are. D-score point estimates move only slightly but standard errors can differ substantially, so inferences drawn from them need re-checking. Backward compatibility is preserved through algorithm = "1.8.7", the pattern this package uses for every break.

    View source ↗
  6. 3y ago

    dscore 1.7.0

    Repairs an LF item-ordering error introduced in 2022 that invalidated the gsed2206 and gsed2208 keys and any analysis relying on LF labels, and introduces gsed2212 as the corrected default. Note that this entry is stamped three minutes after 1.8.0 despite carrying the lower version number, so feed order here does not track release order.

    View source ↗