ACCMV

ACCMV implements inference under the available complete-case missing value assumption for nonmonotone missing data. The repository contains matched R and Python packages with IPW, regression-adjustment, and multiply-robust estimation; exponential-tilt sensitivity analysis; marginal-regression weights; and nonparametric bootstrap confidence intervals.

The estimators support a single primary variable and two-primary-variable average, product-moment, and joint-distribution targets. Missing values are represented by NA in R and numpy.nan in Python.

Installation

Version 0.1.1 has been submitted to CRAN and is awaiting CRAN review. Until it appears in the CRAN package index, install the validated release from GitHub:

# install.packages("remotes")
remotes::install_github("mathcg/ACCMV", subdir = "r")

python -m pip install accmv

Quick start

library(accmv)
fit <- estimate_accmv_single(x, y, method = "mr",
                             n_boot = 499, seed = 1)
fit

from accmv import estimate_single

fit = estimate_single(x, y, method="mr", n_boot=499, random_state=1)
print(fit.estimate, fit.conf_int)

The paper's marginal linear model is available in both languages through fit_accmv_regression() (R) and fit_ipw_regression() (Python). Both use one-based primary-variable column indices so equivalent calls have the same arguments.

See docs/METHODS.md for the API-to-paper mapping and validation/README.md for reproducibility checks. The original research scripts and diabetes demonstration remain at the repository root for provenance.

Reference

If you use this software or the ACCMV method, please cite:

Cheng, G., Chen, Y.-C., Smith, M. A., and Zhao, Y.-Q. (2022). “Handling Nonmonotone Missing Data with Available Complete-Case Missing Value Assumption.” arXiv:2207.02289.

License

MIT © Gang Cheng.