Hourly step recommendations to achieve daily goals for working and older adults

The study addresses how daily incentive thresholds (5000, 7500, 10,000 steps) influence within-day accumulation of steps and how participants with different demographics progress toward daily goals. With high global uptake of wearables and evidence that incentives can increase physical activity, the research seeks to fill a gap in understanding intra-day behavioural responses to incentivised thresholds outside of tightly controlled trials. The purpose is to inform timing and content of personalised prompts to optimise achievement of daily step goals among working-age and older adults. This has public health importance given the burden of physical inactivity and the scalability of tracker-based interventions.

The paper situates its work within literature showing dose–response benefits of physical activity for chronic disease prevention and the use of financial and digital interventions to increase activity. Prior studies largely used aggregated daily steps; fewer examined intra-day patterns, and fewer still analysed behaviour around incentivised thresholds. Evidence suggests prompts can change step behaviour, incentives can shape goal attainment, and older versus working adults differ in diurnal activity patterns. The study builds on NSC3 evaluations of reach and impact by analysing hourly trajectories relative to thresholds.

Design and setting: Retrospective observational analysis of NSC3, a nationwide programme in Singapore (28 Oct 2017–31 Mar 2018). Intra-day recording began Jan 2018; analysis window was 8 Jan–31 Mar 2018. Participants: Subset of NSC3 participants not in the Personal Pledge (more representative of general population), aged ≥17, with valid demographics (no missing age; weight 30–300 kg; height 101–220 cm). Devices included HPB Careeach tracker and inbuilt smartphone accelerometers (Apple HealthKit, Samsung), among others. Data: 30-min time blocks of step counts were cleaned (removed implausible blocks ≥48; negative steps). If any non-zero in a day, remaining empty blocks were imputed as zero, then aggregated to hourly counts. Unit of analysis was participant-day (cross-sectional), yielding 3075 participants and 52,346 participant-days of 24-h data. Incentive structure: Daily thresholds at 5000, 7500, and 10,000 steps earning 10, 25, and 40 HealthPoints, respectively. Outcomes and covariates: Primary outcome was hourly step trajectory. Demographics: sex, age (17–29, 30–39, 40–49, 50–59, ≥60), BMI categories using Asian cutoffs (<18.5, 18.5–22.9, 23–27.4, ≥27.5 kg/m²). Statistical analysis: Descriptive statistics for daily steps by demographic strata; differences tested via linear regression with cluster-robust (by participant) standard errors and joint F-tests. Visualised histograms and hourly means/zero proportions by day-of-week; cumulative hourly trajectories. Two-part model for each hour, stratified by weekdays vs weekends: (1) Zero-part: logistic regression for probability of any steps; (2) Positive-part: gamma regression with log link for positive step counts. For 12 a.m.–12 p.m., covariates were age group, BMI group, sex. For 12 p.m.–12 a.m., models additionally incorporated cumulative steps up to the previous hour, stratified into four intervals tied to incentives (0–4999, 5000–7499, 7500–9999, ≥10,000) and used a scaled continuous cumulative-steps covariate (sum/10,000). Cumulative steps were not used as covariates before noon due to low counts and low prevalence above 5000 by noon. Bayesian estimation: Uninformative priors; MCMC via RStan with four parallel chains, 5000 burn-in iterations each; posterior sample size 20,000 after merging. Convergence assessed with Gelman–Rubin Rhat (target <1.01). Posterior predictive checks compared quantiles (10–25–75–90th) of data vs model. All other analyses conducted in R 3.6.3.

Sample: 3075 participants; mean age 44.2 (SD 13.9); 40.4% male; mean BMI 25.3 (SD 4.5). Total 52,346 participant-days. Daily steps: Mean 10,400 (SD 6100). Distribution: 15.3% <5000 steps; 12.7% 5000–7499; 12.4% 7500–9999; 59.5% ≥10,000. Strong spikes at incentive thresholds, especially at 10,000: days with 9500–9999 steps were 2.08%, increasing sharply to 14.3% at 10,000–10,499. Older adults (≥60) had more days ≥10,000 steps (72.5%) than ages 30–59 (57.2%). By age, mean daily steps increased with age group: 17–29: 9300 (SD 5100); 30–39: 9700 (SD 6200); 40–49: 10,500 (SD 6400); 50–59: 10,900 (SD 6500); ≥60: 11,100 (SD 5100) (p=6.99×10⁻¹⁰). By BMI: underweight 10,000 (SD 4300); normal 10,600 (SD 6800); overweight 10,900 (SD 6400); obese 9600 (SD 5200) (p=1.06×10⁻⁶). By sex: male 11,000 (SD 7200) vs female 10,100 (SD 5400) (p=8.37×10⁻⁴). Hourly patterns: Weekday peaks at 7–9 a.m. (mean 1230, SD 1420), 12–2 p.m. (1300, SD 1260), 6–8 p.m. (1610, SD 1730). Combined, these 6 h delivered ~4140 steps (SD 2800). Weekend peaks at 9–11 a.m. (1320, SD 1790) and 5–7 p.m. (1310, SD 1650), totaling ~2620 steps (SD 2460). Zero-step proportions highest 3–5 a.m.; lowest 1–3 p.m.; higher on weekends (51.2%) than weekdays (45.1%). On weekdays, ≥60 had higher hourly mean steps than ages 30–59 from start until 6 p.m., then the reverse in evenings; ≥60 also had lower zero-step proportions earlier in the day. Model fit: Posterior predictive distributions closely matched empirical quantiles, with pronounced late-day peak at 10,000 steps; better fit on weekdays (more data). Evening step targets and probabilities to reach 10,000: Weekdays—6000 by 6 p.m. → P=0.692; 7000 by 7 p.m. → P=0.910; 8000 by 8 p.m. → P=0.982; 8500 by 9 p.m. → P=0.700; 9500 by 10 p.m. → P=1.000. Weekends—6500 by 6 p.m. → P=0.852; 7000 by 7 p.m. → P=0.583; 8000 by 8 p.m. → P=0.947; 8500 by 9 p.m. → P=0.717; 9500 by 10 p.m. → P=1.000. Table 3 provides stratified evening targets by age, BMI, and sex; generally, participants ≥60 required fewer evening steps than ages 17–59.

Findings indicate that incentive thresholds substantially shape within-day behaviour, particularly the strong clustering at 10,000 steps, suggesting participants exert extra effort to cross the highest reward threshold. Distinct diurnal patterns align with commuting and lunch times on weekdays, accounting for over 40% of steps needed for 10,000 during 25% of the day—prime windows for interventions. Older adults tend to accumulate more steps earlier in the day, while working-age adults accumulate more later, supporting timing of personalised prompts (earlier for older/retired adults; post-work or pre-lunch for working adults). The two-part, history-dependent model provides actionable evening targets to help participants assess progress and likelihood of meeting goals. Differences by BMI and sex suggest nuanced tailoring: e.g., some BMI strata had lower expected remaining steps on weekends; sex differences attenuated when evening steps exceeded 5000. Overall, leveraging real-time tracker data to deliver time- and profile-specific nudges could improve goal attainment and potentially shift population physical activity distributions.

Using high-resolution intra-day data from a nationwide incentive programme, the study quantifies how step goals influence hourly behaviour and provides specific evening step targets to maximise the chance of reaching daily goals. Older adults had earlier-day activity, while working-age adults peaked around commute and lunch times. The modelling framework offers a basis for real-time, personalised prompts via wearables. Future work should experimentally test prompt timing and content, incorporate additional covariates (e.g., occupation), and evaluate generalisability across contexts and alternative activity goals aligned with current guidelines (e.g., minutes of moderate-to-vigorous activity).

Observational design with self-selection into NSC3 limits causal inference; no control group and potential residual confounding. Lack of key covariates (e.g., occupation type) that likely influence step patterns. Analysis focused on intra-day patterns rather than changes in total daily steps over time. Generalisability may be limited outside Singapore due to climate, urban transport, and car ownership patterns that can elevate walking. The arbitrariness of the 10,000-step goal is debated; later NSC seasons shifted toward WHO-aligned time-based activity recommendations. Difficulty increasing activity among participants with very low evening steps suggests underlying health constraints for some individuals. Model fit was better for weekdays than weekends due to fewer weekend days in the analysis window.