Statistical considerations in the analysis of accelerometry-based activity monitor data

Abstract

We review and discuss three statistical aspects of accelerometer-based estimates of physical activity energy expenditure (PAEE):

1) the nature of the relationship between accelerometer output and PAEE;

2) statistical aspects of calibration studies; and

3) two specialized statistical methods that are applicable to the problem of measurement error modeling and missing data methods.

We call for a continuing development of statistical methods that use more characteristics of the accelerometer signal to estimate PAEE, advocate the use of bias and SE statistics and inclusion of cross-validation in accelerometer research designs, and encourage more efforts to understand systematic and random errors in accelerometer-based estimates of PAEE.