← Back to all sparks
N

nmar

ANALYTICS
Velocity0.0

R package nmar by ncn-foreigners — release notes from GitHub.

NMAR landed on CRAN with two nonresponse estimators behind one interface, then started tuning it.

survey statisticsnonresponseempirical likelihoodbootstrapr packagecran
Current state
Three releases in seven weeks, starting from nothing. The initial CRAN release implements empirical likelihood (Qin, Leung and Shao 2002) and both parametric and nonparametric exponential tilting (Riddles, Kim and Im 2016) for estimating means under nonignorable nonresponse, all reachable through a single nmar() call with formula syntax and direct support for survey.design objects. Since then the work has been operational: a configurable bootstrap backend and stricter input validation.
Where it's heading
The package is positioning itself as the general interface to nonignorable-nonresponse estimation rather than a reference implementation of one paper — shared architecture across engines, one formula API, and integration with the survey package so weights and stratification come for free. The follow-up releases suggest the next constraint is compute: bootstrap variance estimation is the expensive part, and it now dispatches to future.apply when a parallel plan exists.
Prediction
Expect further engines under the same nmar() interface or wider bootstrap support, since the architecture was explicitly refactored to share structure across estimators.

Recent moves

  1. 6mo ago

    Bootstrap backend now parallel-aware and configurable

    The bootstrap backend becomes configurable via options(nmar.bootstrap_apply), defaulting to future.apply when a multi-worker plan is active and base lapply otherwise. Exponential tilting also now rejects non-finite covariates and outcomes outright.

    View source ↗
  2. 7mo ago

    CRAN submission fixes and DOI references

    Roxygen return keywords on S3 methods and DOI references added to DESCRIPTION. Housekeeping for the CRAN submission.

    View source ↗
  3. 8mo ago

    NMAR 0.1.0

    ⚡ SPARK

    The first CRAN release establishes the whole package: empirical likelihood and exponential tilting estimators for nonignorable nonresponse, unified behind one nmar() formula interface with survey design support and bootstrap variance estimation.

    View source ↗