Title: An R Package for bias reduction with LogF(1,1) penalty under the MAR mechanism
Authors: Muna AL-Shaaibi - Sultan Qaboos University (Oman) [presenting]
Ronald Wesonga - Sultan Qaboos University (Oman)
Abstract: When data are missing at random (MAR), bias in estimated logit model parameters is inevitable. Although most studies proceed to apply and even publish such results, they are usually misleading and get worse with an increased proportion of missingness. We propose an R package for the bias reduction method, originating from our current study, which explores the penalization of the log-likelihood with the LogF(1,1) penalty. The four main functions with different purposes, including; filling the missing data, applying the Expectation-Maximization (EM) by the method of weights for the missing data, fitting the penalized binomial model under missingness, and summarizing the model output will be presented. The package has been validated using real-life COVID-19 data and some results are discussed.