Lynn McCleary1. 1. Kunin-Lunenfeld Applied Research Unit, Baycrest Centre for Geriatric Care, Toronto, Ontario, Canada. lmccleary@klaru-baycrest.on.ca
Abstract
BACKGROUND: Sample loss and missing data are inevitable in multivariate and longitudinal research. Ad hoc approaches such as analysis of incomplete data or substituting the group mean for missing data, while common, may unnecessarily reduce statistical power and threaten study validity. Multiple imputation for missing data is a newly accessible, methodologically rigorous approach to dealing with the problem of missing data. APPROACH: To (a) discuss the problem of missing data in clinical research, and (b) describe the technique of multiple imputation. A case of analysis of multivariate psychosocial data is presented to illustrate the practice of multiple imputation. RESULTS: The advantages of multiple imputation are it (a) results in unbiased estimates, providing more validity than ad hoc approaches to missing data; (b) uses all available data, preserving sample size and statistical power; (c) may be used with standard statistical software; and, (d) results are readily interpreted. DISCUSSION: Accessible, user-friendly computer programs are available to perform multiple imputation for missing data making ad hoc approaches to missing data obsolete.
BACKGROUND: Sample loss and missing data are inevitable in multivariate and longitudinal research. Ad hoc approaches such as analysis of incomplete data or substituting the group mean for missing data, while common, may unnecessarily reduce statistical power and threaten study validity. Multiple imputation for missing data is a newly accessible, methodologically rigorous approach to dealing with the problem of missing data. APPROACH: To (a) discuss the problem of missing data in clinical research, and (b) describe the technique of multiple imputation. A case of analysis of multivariate psychosocial data is presented to illustrate the practice of multiple imputation. RESULTS: The advantages of multiple imputation are it (a) results in unbiased estimates, providing more validity than ad hoc approaches to missing data; (b) uses all available data, preserving sample size and statistical power; (c) may be used with standard statistical software; and, (d) results are readily interpreted. DISCUSSION: Accessible, user-friendly computer programs are available to perform multiple imputation for missing data making ad hoc approaches to missing data obsolete.
Authors: Nicole Zarrett; Lauren H Law; Dawn K Wilson; Michelle Abraczinskas; Stephen Taylor; Brittany S Cook; Alex Roberts Journal: J Behav Med Date: 2021-03-07
Authors: Pilar Navarro; John Mehegan; Celine M Murrin; Cecily C Kelleher; Catherine M Phillips Journal: Int J Obes (Lond) Date: 2020-08-22 Impact factor: 5.095
Authors: Sandra Cortés; Cinthya Leiva; María José Ojeda; Natalia Bustamante-Ara; Wanjiku Wambaa; Alan Dominguez; Carlos Pasten Salvo; Camila Rodriguez Peralta; Bárbara Rojas Arenas; Diego Vargas Mesa; Ericka Ahumada-Padilla Journal: Environ Health Insights Date: 2022-06-27