A computationally efficient sequential regression imputation algorithm for multilevel data

ABSTRACTDue to the computational burden, especially in high-dimensional settings, sequential imputation may not be practical. In this paper, we adopt computationally advantageous methods by sampling the missing data from their perspective predictive distributions, which leads to significantly improved computation time in the class of variable-by-variable imputation algorithms. We assess the computational performance in a comprehensive simulation study. We then compare and contrast the performance of our algorithm with commonly used alternatives. The results show that our method has a significant advantage over the commonly used alternatives with respect to computational efficiency and inferential quality. Finally, we demonstrate our methods in a substantive problem aimed at investigating the effects of area-level behavioral, socioeconomic, and demographic characteristics on poor birth outcomes in New York State among singleton births.KEYWORDS: Sequential regression imputationmultilevel datacomputational efficiencyfast variable by variable imputationmultiple imputation by chained equations AcknowledgmentsWe thank Dr. Tabassum Insaf for providing assistance in accessing the New York State Vital Records Registry data.Disclosure statementNo potential conflict of interest was reported by the author(s).