Akademik Veri Yönetim Sistemi
International Biometric Conference, Barcelona, İspanya, 08 Temmuz 2018 - 13 Temmuz 2021, ss.99-100
Among many analytical challenges, presence of incomplete categorical variables is an additional complexity to the analysis of multilevel data. Inference by multiple imputation (MI) can offer a statistically-sound solution inference, however, model-based MI routines can quickly become problematic even with a moderately high number of categorical variables. To provide an alternative solution to this issue, we develop computationally feasible routines for conducting inference by multiple imputation for ordinal variables using the notion of calibration. Particularly, we propose rounding rules to be used with the computationally efficient sequential regression imputation (CESRI) method (Akkaya-Hocagil and Yucel 2017). These rules allow practitioners to obtain usable set of imputations and hence facilitate the inferentially sound technique of multiple imputation. Through simulation results, we show that our methods lead to estimates with sound inferential quality. We also illustrate our techniques in a data application.