Who Are We Excluding From Physical Activity Research? Examining the Potential for Exclusion Bias in Pedometer Data Processing

Abstract

Background: Pedometers are a useful measure of physical activity (PA) but whether systematic bias exists when using differing inclusion criterion for pedometer-derived PA data is unknown. We undertake an exploration of previously published criteria for pedometer data cleaning and examine the impact of different inclusion criteria on sample size retention and participant exclusion by key sociodemographic characteristics.

Methods: Data were drawn from a community survey in Aotearoa/New Zealand. Sociodemographic information and self-reported PA were collected via face-to-face surveys; participants were asked to wear a Yamax CW300 pedometer for 7 days. Analyses involved removing extreme outliers and determining minimum steps per day for inclusion, determining whether day 1 removal was required, examining risk of bias using different inclusion criteria, and examining convergent and concurrent validity of criteria with the lowest bias potential.

Results: Pedometer data were available for 895 participants. A threshold of 100 steps/day was deemed appropriate to define a valid day. All days were correlated with each other; intraclass correlation coefficients were low and did not change meaningfully with removal of the first day of data. Participant retention reduced, and bias in participant inclusion increased, with increasing stringency of data inclusion criterion applied. Evidence for convergent and concurrent validity in the 2 models with the lowest risk of exclusion bias was demonstrated.

Conclusion: Increasing stringency in pedometer data inclusion criteria can result in significant and biased loss of sample size. Clear reporting of data cleaning methods and rationale (including consideration of potential for bias) is needed in pedometer-based PA research.

Keywords: data treatment; equity; measurement; pedometry; steps.