Age-Dependent Heterogeneity in the Association Between Physical Activity and Mental Distress: A Causal Machine Learning Analysis of 3.2 Million U.S. Adults
Yuan Shan · 2026
Read Age-Dependent Heterogeneity in the Association Between Physical Activity and Mental Distress: A Causal Machine Learning Analysis of 3.2 Million U.S. Adults by Yuan Shan (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-19066.txt.
Introduction
Shan asks whether activity's link to frequent distress, defined as 14 or more days a month, varies by age. Ten U.S. survey waves covering more than 3 million adults show a clear age gradient in weighted logistic regression. The adjusted odds ratio is 0.89 for young adults 18 to 24 and strengthens steadily with age. Over the decade, the young adult effect eroded to null in 2018 and 2024. Causal Forest names age as the top driver of treatment effect heterogeneity. Sensitivity checks confirm this, suggesting the exercise link may not generalize to youngest adults.
Catalog record
| CREATOR | Yuan Shan |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 24660 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.19066
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Age-Dependent Heterogeneity in the Association Between Physical Activity and Mental Distress: A Causal Machine Learning Analysis of 3.2 Million U.S. Adults Yuan Shan Abstract Physical activity (PA) is widely recognized as protective against mental distress, yet whether this benefit varies systematically across population subgroups remains poorly understood. Using pooled data from ten consecutive annual waves of the U.S. Behavioral Risk Factor Surveillance System (2015–2024; n = 3,242,218 n=3{,}242{,}218 ), we investigate heterogeneity in the association between leisure-time PA and frequent mental distress (FMD, ≥ \geq 14 days/month) across age groups. Survey-weighted logistic regression reveals a striking age gradient: the adjusted odds ratio for PA ranges from 0.89 among young adults (18–24) to 0.50 among adults aged 55–64, with the protective association strengthening monotonically with age. Temporal analysis across all ten years shows that the young-adult PA effect has been eroding over the past decade , with the 18–24 OR reaching 1.01 (null) in both 2018 and 2024—paralleling the deepening youth mental health crisis. Causal Forest via Double Machine Learning independently identifies age as the dominant driver of treatment effect heterogeneity (feature importance = 0.39, 2.5 × \times the next predictor). E-value sensitivity analysis, propensity score overlap checks, placebo tests, and imputation comparisons confirm the robustness of the findings. These results suggest that the well-documented exercise–mental health link may not generalize to the youngest adult population, whose distress appears increasingly driven by stressors that PA alone cannot mitigate. Keywords: physical activity, mental distress, heterogeneous treatment effects, causal forest, BRFSS, age moderation 1 Introduction Mental health disorders represent a leading public health challenge in the United States, affecting more than one in five adults and imposing an economic burden exceeding $280 billion annually [ 3 ] . A widely used population-level metric is frequent mental distress (FMD), defined as self-reporting \geq 14 days of poor mental health in the preceding 30 days. FMD captures a broad spectrum of psychological burden—including stress, depression, and emotional difficulties—and is robustly associated with increased healthcare utilization, reduced work productivity, and diminished quality of life. The youth mental health crisis. FMD prevalence has risen sharply over the past two decades, with the most pronounced increases among younger adults. 23 documented that serious psychological distress among 18–25-year-olds increased by 71% between 2008 and 2017, a trend attributed primarily to cohort effects rather than period effects. Subsequent analyses confirmed that this trajectory continued through 2020, with young adults experiencing the sharpest declines in mental wellbeing [ 24 , 8 ] . The drivers of this generational shift remain debated but likely include the rise of smartphone-based social media, increased academic and economic pressure, and post-pandemic adjustment difficulties [ 9 , 1 ] . Physical activity as a modifiable protective factor. Among behavioral interventions, physical activity (PA) is one of the most promising and well-studied. A large body of evidence from meta-analyses of prospective cohort studies [ 20 , 16 ] and intervention trials [ 21 ] demonstrates that regular PA reduces the incidence and severity of depression and anxiety. Neurobiological mechanisms include endorphin and endocannabinoid release, hypothalamic-pituitary-adrenal (HPA) axis regulation, reduced systemic inflammation, improved sleep architecture, and enhanced neuroplasticity [ 7 , 13 ] . The World Health Organization recommends at least 150 minutes of moderate-intensity aerobic activity per week [ 27 ] , yet approximately 25% of U.S. adults remain physically inactive. The gap: age-specific effect heterogeneity. The landmark study by 5 , analyzing 1.2 million BRFSS respondents (2011–2015), reported that individuals who exercised had 43% fewer days of poor mental health per month and concluded that the benefit was “consistent” across demographic subgroups, including age. However, that analysis controlled for age via matching rather than testing for formal effect modification. Several smaller studies have noted that the PA–mental health association may differ by age [ 10 , 14 , 11 ] , but no large-scale analysis has systematically quantified age-dependent heterogeneity in the PA–FMD relationship using nationally representative data across multiple years. Contributions. This study makes three contributions: 1. We update and extend the PA–FMD association using pooled data from 3.2 million adults across ten consecutive BRFSS annual waves (2015–2024), conducting the first systematic age-stratified analysis to reveal a dramatic, monotonically increasing protective gradient from ages 18–24 (OR = 0.89) to 55–64 (OR = 0.50). 2. We document a novel temporal finding: the already-weak protective association among 18–24-year-olds has been eroding over the past decade , reaching null in both 2018 and 2024. 3. We apply causal machine learning—specifically, Causal Forest via Double Machine Learning [ 26 , 6 ] —to independently confirm that age is the dominant driver of treatment effect heterogeneity in a fully data-driven, nonparametric framework. 2 Data 2.1 Source and Design The Behavioral Risk Factor Surveillance System (BRFSS) is a CDC-administered cross-sectional telephone survey of non-institutionalized U.S. adults ( \geq 18 years) conducted annually across all 50 states, the District of Columbia, and participating territories [ 4 ] . The BRFSS employs disproportionate stratified sampling (DSS) for landline telephones and random sampling for cellular telephones, with iterative proportional fitting (raking) applied to produce population-representative weights [ 17 ] . We pooled all ten consecutive annual BRFSS waves from 2015 through 2024, spanning a full decade of dramatic shifts in mental health epidemiology while using consistent core questionnaire items. The raw pooled dataset contained 4,410,255 respondents. After applying inclusion criteria (valid responses on all study variables) and complete-case analysis, the final analytic sample comprised 3,242,218 adults (85.2% retention). 2.2 Variables The outcome is FMD, derived from the BRFSS item MENTHLTH : “For how many days during the past 30 days was your mental health not good?” Responses of 88 (“none”) are recoded to 0; responses of 77 (don’t know) and 99 (refused) are excluded. FMD is defined as \geq 14 days. The exposure is leisure-time PA, from the calculated variable _TOTINDA : whether the respondent engaged in any physical activity or exercise during the past 30 days other than their regular job (1 = active, 2 = inactive). Covariates include: sex (2 levels), age group (6 categories: 18–24, 25–34, 35–44, 45–54, 55–64, 65+), race/ethnicity (5 groups: White NH, Black NH, Other NH, Multiracial NH, Hispanic), education (4 levels: < high school through college graduate), household income (5 harmonized levels across survey years: < $15K through $50K+), and BMI category (4 levels: underweight, normal, overweight, obese). Survey weights ( _LLCPWT ) are divided by 10 for pooled estimates. 2.3 Missing Data The primary source of missingness was income (14.8% of pooled observations after recoding). Income variable coding changed between BRFSS waves: pre-2021 waves used a 5-level grouping ( _INCOMG ), while 2021+ used a 7-level grouping ( _INCOMG1 ). We harmonized to a common 5-level scale by collapsing upper income categories. The main analysis uses the complete-case sample; sensitivity analyses compare results under conditional random imputation stratified by age and education (Section 3.4 ). 3 Methods Figure 1 provides an overview of the analytical pipeline. Figure 1: Analytical pipeline. From the pooled BRFSS data (top), three parallel analysis streams converge on the central finding (bottom): the confirmatory arm provides precise age-stratified estimates; the causal ML arm independently discovers age as the dominant heterogeneity driver; the robustness arm validates assumptions. 3.1 Survey-Weighted Logistic Regression We model