Measurement Days Reduce Bias While Sample Size Improves Precision in Behavior Comparisons

A bootstrap simulation study evaluating 24-hour physical activity monitoring in office workers demonstrates that the duration of data collection dictates measurement trueness, whereas participant sample size and group balance govern statistical precision. When comparing movement behaviors between distinct populations, measuring participants for too few days introduces systematic bias into the estimated group difference, while enrolling larger numbers of participants does not eliminate this single-day distortion. The findings establish that achieving an accurate assessment of differences in physical behavior requires balancing the number of monitored working days against overall participant recruitment.

Researchers examined the methodological trade-offs inherent in physical behavior surveillance, where constrained resources frequently force investigators to choose between tracking more individuals or tracking individuals across longer timeframes. Insufficient data collection can generate inaccurate and statistically uncertain group differences, while excessive monitoring risks misallocating institutional resources. To resolve these operational trade-offs, the research team evaluated study performance across multiple sampling configurations, defining accuracy as the combined influence of trueness, representing the closeness of an estimate to the reference value, and precision, reflecting the width of the corresponding confidence interval.

The investigation utilized an empirical parent dataset comprising 103 overweight office workers with a body mass index of 25 kilograms per square meter or greater and 122 normal-weight office workers with a body mass index below 25 kilograms per square meter. Each participant wore an accelerometer affixed to the thigh continuously over a seven-day monitoring protocol. This measurement protocol recorded 24-hour continuous time expenditure across four distinct behavioral categories: sitting, standing, moving, and sleeping.

Using bootstrap resampling techniques applied to this parent dataset, the researchers simulated 96 distinct data collection strategies formed by crossing eight participant sample sizes, four measurement durations, and three group allocation ratios. The simulated total sample sizes comprised 18, 36, 54, 72, 90, 108, 126, and 144 office workers. The monitoring durations evaluated included 1, 2, 3, and 4 working days per participant, while the sample compositions incorporated balanced 1:1 allocations alongside unbalanced 1:2 and 2:1 ratios between overweight and normal-weight individuals.

The simulation results revealed that trueness and precision respond to distinct design parameters. Increasing the total number of participants within a study did not modify the estimated group difference at any given number of monitored working days. Restricting data collection to a single working day introduced substantial bias into the estimated behavioral difference, whereas monitoring across two, three, or four working days generated estimates that progressively converged toward the observed reference difference from the complete seven-day parent dataset. In contrast, precision, defined by the width of the 95 percent confidence interval, improved systematically as both the number of participants and the number of monitoring days increased.

Study balance also played a critical role in statistical precision and overall analytical performance. For any fixed quantity of total measurements, balanced designs featuring an equal 1:1 allocation of overweight and normal-weight workers generally yielded more precise group difference estimates than unbalanced 1:2 or 2:1 configurations. Furthermore, the likelihood of detecting a statistically significant difference between groups was driven jointly by the total number of participants and the number of monitoring days per individual. The authors concluded that study planners must explicitly weigh these design trade-offs, alongside the financial and logistical costs of recruiting additional participants versus extending monitoring days within each target group.

Methodological benchmarking across the 96 simulation models directly contrasted each subsampling strategy against the complete seven-day parent dataset. In compositional space, trueness was evaluated by measuring the size of the deviation in group differences relative to the empirical baseline, while precision was indexed by the width of the corresponding 95 percent confidence intervals. Tracking the probability of obtaining statistically significant group differences provided an integrated assessment of overall study performance, capturing the combined consequences of trueness and precision across diverse monitoring configurations.

These methodological relationships highlight that investigator efforts cannot treat participant recruitment and monitoring duration as interchangeable parameters. Because trueness is determined solely by the number of working days monitored rather than sample size, enrolling larger cohorts cannot correct the bias introduced by single-day assessments. Instead, achieving accurate group comparisons requires ensuring adequate measurement days per individual to secure trueness, while jointly tuning sample size, monitoring days, and group balance to achieve adequate precision without expending excess data collection resources.

Subject of Research: Accuracy and study design trade-offs in 24-hour physical behavior measurement

Article Title: Accuracy of a difference in 24-hour physical behaviors between two groups: effects of balanced and unbalanced group sizes across scenarios with varying numbers of measurement days

Article References: Brusaca, L. A., Mathiassen, S. E., Gupta, N., Oliveira, A. B., & Hallman, D. M. (2026). Accuracy of a difference in 24-hour physical behaviors between two groups: effects of balanced and unbalanced group sizes across scenarios with varying numbers of measurement days. Journal of Activity, Sedentary and Sleep Behaviors. https://doi.org/10.1186/s44167-026-00124-z

Image Credits: AI Generated

DOI: 10.1186/s44167-026-00124-z

Keywords: 24-hour physical behaviors, accelerometry, trueness, precision, measurement bias, bootstrap resampling, compositional data analysis, office workers, sedentary behavior, sitting time, standing time, sleep tracking, body mass index, study design, balanced group sizes, confidence interval width, sample size