Missing Value Imputation for Physical Activity Data Measured by Accelerometer

• Activity Number: 184
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Sports
• Author(s): JungAe Lee*
• Companies: Washington University in St. Louis
• Keywords: accelerometer data ; missing data ; zero-inflated model ; multivariate poisson model ; physical activity

Abstract

An accelerometer, a small device to wear on the hip or wrist, is becoming a promising tool in epidemiological and clinical studies to measure the physical activity objectively. The data set consists of a series of activity counts at every 15s, 30s or 60s epochs, and displays a person's activity pattern throughout a day, Unfortunately, the collected data can imply irregular missing intervals because of incompliance of participants, and often make the data useless. In this study, we propose a method to predict the missing counts in the non-wearing periods. The imputation is performed at minute level, and based on a multivariate zero-inflated Poisson regression. Including an autoregressive term is a main feature of this model because it allows us to account for the serial nature of activity counts minute by minute. Functional analysis of variance is also utilized to see how demographic characteristics are associated with physical activities over time. We demonstrate the performance of this method through a real data example from 2003-2004 National Health and Nutrition Examination Survey data.