Ilmingutest

Ilmingutest is a statistical framework for evaluating the robustness of imputation methods in datasets with missing values. It aims to quantify how the choice of imputation model influences downstream analyses and to produce a composite score, the Ilmingutest score, that captures stability, accuracy, and distributional fidelity of imputed data.

Method: The procedure selects a missingness mechanism and several imputation models. For each model, multiple imputations generate several completed datasets. Imputed values are compared across imputations to assess stability, using metrics such as imputation variance and agreement. Each dataset is used to train a downstream model, whose performance is compared to a reference or ground truth. Results are aggregated into the Ilmingutest score. Higher scores indicate greater robustness.

Applications: It is used in methodological research to benchmark imputation methods and in teaching to illustrate how missing data handling affects results. It can also guide data-sharing workflows by selecting imputations that better preserve analytic validity.

Limitations: The score depends on the chosen imputations, downstream tasks, and evaluation metrics; it may be sensitive to sample size and the missingness mechanism; and computing multiple imputations across several models can be resource-intensive.

See also: multiple imputation, missing data, data imputation, model validation, data quality metrics.

Further reading: standard texts on multiple imputation and data quality provide background on the concepts underlying imputation evaluation.