Home/Selected work/NHANES sedentary behavior

Public health · Wearable sensors · 2025—present

NHANES Sedentary Behavior Estimation

Estimating how much waking time people in the United States spend sedentary by combining wrist accelerometry, deep-learning posture labels, sleep detection, and complex survey inference.

RPythonNHANESDeep learningSurvey inference
PopulationNHANES 2011–2014
Analytic sample14,608 participants
Primary estimate11.35 h/day
StatusManuscript in preparation

How sedentary is the U.S. population?

A wrist accelerometer records movement continuously, but it does not directly say whether a person is sitting, sleeping, standing, or not wearing the device.

This project builds those behavioral labels before estimating waking sedentary time. The analysis uses NHANES 2011–2014 because it pairs nationally representative sampling with week-long wrist accelerometry. Survey weights then turn participant-level estimates into population-level results.

My role

Built the analysis workflow, joined posture and sleep predictions, derived participant-level sedentary measures, implemented survey-weighted estimation, and produced the manuscript tables and figures.

From 80 Hz signals to hours per day.

01

Prepare signals

Align raw wrist acceleration with participant records and quality-control valid wear days.

02

Classify posture

Use CHAP deep-learning predictions to identify sitting or lying posture at the epoch level.

03

Identify sleep

Combine SWaN sleep/wear predictions with a calibrated cutoff informed by self-reported sleep.

04

Aggregate

Convert epoch labels into hourly and participant-level waking sedentary time.

05

Estimate

Apply NHANES survey weights, strata, and primary sampling units to obtain nationally representative means and confidence intervals.

Workflow from raw wrist accelerometry through posture and sleep classification to survey-weighted sedentary estimates
Analysis workflow from raw NHANES wrist accelerometry to population estimates. Source: project repository.

Waking sedentary time averaged 11.35 hours per day.

Waking sedentary time11.35 h

95% CI: 11.26–11.44 hours per day

Total waking time17.09 h

95% CI: 16.99–17.19 hours per day

Age pattern13.52 h

Men aged 65+; women aged 65+ averaged 12.89 h/day

The pooled estimate changed modestly across survey cycles: 11.46 hours per day in 2011–2012 and 11.25 hours in 2013–2014. Sedentary time rose with age in both men and women, from roughly 10 hours among participants under 18 to approximately 13 hours among adults 65 and older.

Selected survey-weighted estimates
GroupMen, mean (95% CI)Women, mean (95% CI)
All ages11.48 (11.35, 11.62)11.22 (11.12, 11.33)
Under 1810.54 (10.35, 10.73)10.20 (10.03, 10.36)
45–6411.93 (11.72, 12.14)11.69 (11.51, 11.86)
65 and older13.52 (13.33, 13.72)12.89 (12.62, 13.15)

When sedentary time occurs matters too.

Hourly waking sedentary time across the day
Survey-weighted hourly waking sedentary patterns. Source: final analysis output.
Distribution and 24-hour composition of sedentary, active, and sleep or nonwear time
Distribution of waking sedentary time alongside the 24-hour behavioral composition.

What the estimate does—and does not—mean.

Classification is still measurement.

The final estimate depends on posture and sleep classifiers, valid-wear rules, and the selected sleep calibration cutoff. Survey weights address the NHANES sampling design, but they do not remove classification error. The repository therefore includes cutoff-sensitivity analysis, quality-control outputs, and cycle-specific estimates.

Next case studyUnity GPU Efficiency Analytics