A nonparametric multiple imputation approach for missing categorical data

Abstract

BackgroundIncomplete categorical variables with more than two categories are common in public health data. However, most of the existing missing-data methods do not use the information from nonresponse (missingness) probabilities.MethodsWe propose a nearest-neighbour multiple imputation approach to impute a missing at random categorical outcome and to estimate the proportion of each category. The donor set for imputation is formed by measuring distances between each missing value with other non-missing values. The distance function is calculated based on a predictive score, which is derived from two working models: one fits a multinomial logistic regression for predicting the missing categorical outcome (the outcome model) and the other fits a logistic regression for predicting missingness probabilities (the missingness model). A weighting scheme is used to accommodate contributions from two working models when generating the predictive score. A missing value is imputed by randomly selecting one of the non-missing values with the smallest distances. We conduct a simulation to evaluate the performance of the proposed method and compare it with several alternative methods. A real-data application is also presented.ResultsThe simulation study suggests that the proposed method performs well when missingness probabilities are not extreme under some misspecifications of the working models. However, the calibration estimator, which is also based on two working models, can be highly unstable when missingness probabilities for some observations are extremely high. In this scenario, the proposed method produces more stable and better estimates. In addition, proper weights need to be chosen to balance the contributions from the two working models and achieve optimal results for the proposed method.ConclusionsWe conclude that the proposed multiple imputation method is a reasonable approach to dealing with missing categorical outcome data with more than two levels for assessing the distribution of the outcome. In terms of the choices for the working models, we suggest a multinomial logistic regression for predicting the missing outcome and a binary logistic regression for predicting the missingness probability.Electronic supplementary materialThe online version of this article (doi:10.1186/s12874-017-0360-2) contains supplementary material, which is available to authorized users.

A simulation study on missing data imputation for dichotomous variables using statistical and machine learning methods

“…KNN and RF have been reported to have excellent imputation performance in relevant studies [5][6][7][8][9][10]27 , but these researches were based on real-data applications with continuous or mixed variables in limited application scenarios. In this study, KNN only had a relatively moderate imputation accuracy in scenarios with a value distribution of 7:3, but it had poor performance in most scenarios, which may be caused by the local structure of data in these simulation scenarios.…”

“…Many researchers have successively used different datasets to compare the performance of traditional statistical and machine learning imputation methods, but the conclusions were different. Wei et al 5 , Waljee et al 6 , Shah et al 7 demonstrated respectively that RF outperforms other imputation methods in their datasets; Jerez et al 8 , Zhou et al 9 , Jadhav et al 10 found KNN outperforms other imputation methods in their datasets; Chlioui et al 11 found SVM performs best in two numeric datasets, while Tsai 12 found DT performs best in mixed datasets. Furthermore, both the ensemble learning (EL) algorithm proposed by Wang 13 and the generative adversarial imputation nets (GAIN) algorithm proposed by Dong 14 have been reported as possessing satisfactory imputation performance.…”

Development of Machine Learning Algorithms to Predict Being Lost to Follow-up After Hip Arthroscopy for Femoroacetabular Impingement Syndrome

“… 11 Multiple imputation was applied for variables with less than 30% missing data. 12 , 13 Potential covariates included preoperative demographic variables routinely collected in the secure clinical repository. A total of 27 preoperative demographic variables were included age, body mass index, sex, race, smoking history, alcohol use history, hypertension, diabetes mellitus (type I or II), autoimmune conditions (systemic lupus erythematosus, etc.…”

Weekend delivery and maternal–neonatal adverse outcomes in low‐risk pregnancies in the United States: A population‐based analysis of 3‐million live births

“…Methods of imputations were as follows: predictive mean matching for continuous variables, logistic regression imputation for dichotomous variables, and polytomous regression imputation for categorical variables with more than 2 levels. [24][25][26] Multiple imputation processes were repeated for all variables (1 predictor, 10 outcomes, and 23 covariates) five times to generate five complete, imputed data sets using the "mice" function of the "mice" package 24 in R software. 27 Estimates were calculated by pooling the five sets of results from logistic regression models of five imputed data sets, using "pool" function of the "mice" package.…”