David A Wagstaff1, Sibylle Kranz, Ofer Harel. 1. Department of Human Development and Family Studies, College of Health and Human Development, The Pennsylvania State University, University Park, PA 16802, USA. daw22@psu.edu
Abstract
BACKGROUND: Addressing missing data on body weight, height, or both is a challenge many researchers face. In calculating the body mass index (BMI) of study participants, researchers need to impute the missing data. OBJECTIVE: A multiple imputation through a chained equations approach was used to determine whether one should first impute the missing anthropometric data and then calculate BMI or use an imputation model to obtain BMI. DESIGN: The present study used computer simulation to address the question of how to calculate BMI when there is missing data on weight and height. The simulated data reflected data gathered on non-Hispanic white youths (n = 905) aged 2-18 y, who participated in the 1999-2000 National Health and Nutrition Examination Survey (NHANES). RESULTS: The simulation indicated that it made little difference in the accuracy with which the youths' mean BMIs were estimated when the data were missing completely at random. However, the use of a model to impute BMI was favored slightly when the data were missing at random and the imputation model included the variable used to determine missingness. CONCLUSION: The present findings extend the use of passive imputation and the use of multiple imputation through a chained equations approach to an area of critical public health importance.
BACKGROUND: Addressing missing data on body weight, height, or both is a challenge many researchers face. In calculating the body mass index (BMI) of study participants, researchers need to impute the missing data. OBJECTIVE: A multiple imputation through a chained equations approach was used to determine whether one should first impute the missing anthropometric data and then calculate BMI or use an imputation model to obtain BMI. DESIGN: The present study used computer simulation to address the question of how to calculate BMI when there is missing data on weight and height. The simulated data reflected data gathered on non-Hispanic white youths (n = 905) aged 2-18 y, who participated in the 1999-2000 National Health and Nutrition Examination Survey (NHANES). RESULTS: The simulation indicated that it made little difference in the accuracy with which the youths' mean BMIs were estimated when the data were missing completely at random. However, the use of a model to impute BMI was favored slightly when the data were missing at random and the imputation model included the variable used to determine missingness. CONCLUSION: The present findings extend the use of passive imputation and the use of multiple imputation through a chained equations approach to an area of critical public health importance.
Authors: Honghu Liu; Ron D Hays; John L Adams; Wen-Pin Chen; Diana Tisnado; Carol M Mangione; Cheryl L Damberg; Katherine L Kahn Journal: Health Serv Res Date: 2005-06 Impact factor: 3.402
Authors: R J Kuczmarski; C L Ogden; L M Grummer-Strawn; K M Flegal; S S Guo; R Wei; Z Mei; L R Curtin; A F Roche; C L Johnson Journal: Adv Data Date: 2000-06-08
Authors: Ofer Harel; Emily M Mitchell; Neil J Perkins; Stephen R Cole; Eric J Tchetgen Tchetgen; BaoLuo Sun; Enrique F Schisterman Journal: Am J Epidemiol Date: 2018-03-01 Impact factor: 4.897
Authors: Rameela Raman; Wencong Chen; Michael O Harhay; Jennifer L Thompson; E Wesley Ely; Pratik P Pandharipande; Mayur B Patel Journal: BMC Med Res Methodol Date: 2021-05-06 Impact factor: 4.615