Introduction
Loneliness has gained increasing recognition as a major societal and public health challenge in recent years (Goldman et al., Reference Goldman, Khanna, El Asmar, Qualter and El-Osta2024). Previous research has demonstrated associations between loneliness and a range of adverse health outcomes, most notably depression (Abdellaoui et al., Reference Abdellaoui, Sanchez-Roige, Sealock, Treur, Dennis, Fontanillas and Boomsma2019; Goosby, Bellatorre, Walsemann, & Cheadle, Reference Goosby, Bellatorre, Walsemann and Cheadle2013), severe infections (Elovainio et al., Reference Elovainio, Komulainen, Sipilä, Pulkki-Råback, Alonso, Pentti and Kivimäki2023), cardiovascular disease (Abdellaoui et al., Reference Abdellaoui, Sanchez-Roige, Sealock, Treur, Dennis, Fontanillas and Boomsma2019; Hakulinen et al., Reference Hakulinen, Pulkki-Råback, Virtanen, Jokela, Kivimäki and Elovainio2018), and all-cause mortality (Elovainio et al., Reference Elovainio, Hakulinen, Pulkki-Råback, Virtanen, Josefsson, Jokela and Kivimäki2017; Wang et al., Reference Wang, Gao, Han, Yu, Long, Jiang and Zhao2023; Yu et al., Reference Yu, Cho, Westrick, Chen, Langa and Kobayashi2023). However, the mechanisms underlying these associations remain incompletely understood.
Adverse health behaviors represent a potential mechanism through which loneliness might affect health (Christiansen, Larsen, & Lasgaard, Reference Christiansen, Larsen and Lasgaard2016). The evolutionary theory of loneliness offers a useful theoretical framework, proposing that individuals who feel lonely may display hypervigilance to social threats (Hawkley & Cacioppo, Reference Hawkley and Cacioppo2010; Qualter et al., Reference Qualter, Vanhalst, Harris, Van Roekel, Lodder, Bangee and Verhagen2015). Over time, this sustained alertness may impair executive functioning and self-regulation, thereby heightening susceptibility to impulsive coping strategies and maladaptive health behaviors (Hawkley & Cacioppo, Reference Hawkley and Cacioppo2010). Consistent with this perspective, cohort studies in older adults and genetic evidence demonstrate that loneliness is associated with physical inactivity (Hawkley, Thisted, & Cacioppo, Reference Hawkley, Thisted and Cacioppo2009; Pengpid, Peltzer, & Anantanasuwong, Reference Pengpid, Peltzer and Anantanasuwong2023; Shankar, McMunn, Banks, & Steptoe, Reference Shankar, McMunn, Banks and Steptoe2011), smoking (Shankar et al., Reference Shankar, McMunn, Banks and Steptoe2011; Wootton et al., Reference Wootton, Greenstone, Abdellaoui, Denys, Verweij, Munafò and Treur2021), and underweight (Pengpid et al., Reference Pengpid, Peltzer and Anantanasuwong2023). Among adults experiencing loneliness, such adverse health behaviors might contribute to multiple negative health outcomes, including an increased risk of depression (Liang et al., Reference Liang, Zhou, He, Zhang, Wu, Luo and Zhang2024), severe infections (Elovainio et al., Reference Elovainio, Komulainen, Sipilä, Pulkki-Råback, Alonso, Pentti and Kivimäki2023), cardiovascular disease (Hakulinen et al., Reference Hakulinen, Pulkki-Råback, Virtanen, Jokela, Kivimäki and Elovainio2018; Liang et al., Reference Liang, Zhou, He, Zhang, Wu, Luo and Zhang2024), obesity (Liang et al., Reference Liang, Zhou, He, Zhang, Wu, Luo and Zhang2024), and premature mortality (Elovainio et al., Reference Elovainio, Hakulinen, Pulkki-Råback, Virtanen, Josefsson, Jokela and Kivimäki2017). However, recent cohort studies of older adults that adjusted for a broad range of confounders have found no evidence for links between loneliness and subsequent physical activity (Hong et al., Reference Hong, Nakamura, Berkman, Chen, Shiba, Chen and VanderWeele2023; Kobayashi & Steptoe, Reference Kobayashi and Steptoe2018; Schrempft, Jackowska, Hamer, & Steptoe, Reference Schrempft, Jackowska, Hamer and Steptoe2019), alcohol use (Hong et al., Reference Hong, Nakamura, Berkman, Chen, Shiba, Chen and VanderWeele2023; Kobayashi & Steptoe, Reference Kobayashi and Steptoe2018), smoking (Hong et al., Reference Hong, Nakamura, Berkman, Chen, Shiba, Chen and VanderWeele2023; Kobayashi & Steptoe, Reference Kobayashi and Steptoe2018), and overweight (Hong et al., Reference Hong, Nakamura, Berkman, Chen, Shiba, Chen and VanderWeele2023; Kobayashi & Steptoe, Reference Kobayashi and Steptoe2018). These findings suggest earlier findings may partly reflect bias, such as residual confounding. Moreover, research on loneliness-health behavior linkages predominantly focused on older adults (Hawkley & Capitanio, Reference Hawkley and Capitanio2015; Kobayashi & Steptoe, Reference Kobayashi and Steptoe2018), failing to systematically examine these behavioral precursors of disease and disability across the first half of life.
Adolescence is a crucial developmental period when future adult behaviors are dynamically shaped, and in which loneliness is particularly prevalent (Luhmann & Hawkley, Reference Luhmann and Hawkley2016; von Soest, Luhmann, & Gerstorf, Reference von Soest, Luhmann and Gerstorf2020). However, findings from older adult populations may not directly generalize to earlier life stages, as risky health behaviors such as alcohol use, physical exercise, and smoking in adolescence and emerging adulthood are often socially embedded and shaped by peer norms, social network processes, and identity formation (Montgomery et al., Reference Montgomery, Donnelly, Bhatnagar, Carlin, Kee and Hunter2020), potentially reducing the relative importance of loneliness as an independent driver of these behaviors. Across early life stages, prior research has linked loneliness and lack of social connections to poorer self-rated health and cardiometabolic risk factors such as high blood pressure, elevated body mass index (BMI), cholesterol, and inflammation (Goosby et al., Reference Goosby, Bellatorre, Walsemann and Cheadle2013; Rangul et al., Reference Rangul, Brennsaeter, Eik-Nes, Kvaløy and Friis2025; Yang et al., Reference Yang, Boen, Gerken, Li, Schorpp and Harris2016). A large-scale Danish study further demonstrated associations between loneliness and health behaviors among young people (Jensen et al., Reference Jensen, Friis, Maindal, Hargaard, Knudsen, Grønkjaer and Lasgaard2025); yet its cross-sectional design precluded it from establishing the temporal ordering and directionality of these associations. By contrast, a recent US longitudinal cohort study applying a causal inference framework found no support for causal associations between adolescent loneliness and multiple health behaviors in adulthood, including smoking, alcohol use, physical activity, and overweight/obesity (Kim, Wilkinson, Holt-Lunstad, & VanderWeele, Reference Kim, Wilkinson, Holt-Lunstad and VanderWeele2025). Such causal inference approaches involve adjustment for preexisting levels of the exposure, the outcome, and a wide range of confounders to minimize the risk of reverse causality and residual confounding (Mulder, Usami, & Hamaker, Reference Mulder, Usami and Hamaker2025; VanderWeele, Reference VanderWeele2017, Reference VanderWeele2021), consequently providing more robust insights into potential causal linkages. However, systematic investigations into links between loneliness and adverse health behaviors across the first half of life – from adolescence through emerging adulthood and young adulthood – remain scarce.
The reverse direction has received even less attention, despite prior work suggesting that adverse health behaviors may shape social connections and loneliness. For example, as certain health behaviors such as smoking have become increasingly more stigmatized, individuals who engage in these behaviors face social exclusion and marginalization (Christakis & Fowler, Reference Christakis and Fowler2008; von Soest & Pedersen, Reference von Soest and Pedersen2014). Supporting this notion, a British cohort study demonstrated that smoking was prospectively associated with increased loneliness in older adults (Philip et al., Reference Philip, Bu, Polkey, Brown, Steptoe, Hopkinson and Fancourt2022). Mendelian randomization studies further showed bidirectional loneliness–smoking effects (Wootton et al., Reference Wootton, Greenstone, Abdellaoui, Denys, Verweij, Munafò and Treur2021), as well as the effects of body fat and BMI on loneliness (Abdellaoui et al., Reference Abdellaoui, Sanchez-Roige, Sealock, Treur, Dennis, Fontanillas and Boomsma2019). In contrast, some evidence suggests that some health behaviors may also have protective effects on psychosocial outcomes. A US cohort study employing causal inference methodology showed that moderate alcohol use in young adulthood was associated with a lower risk of depression in midlife (Visontay et al., Reference Visontay, Mewton, Slade, Aris and Sunderland2023), while an Australian household panel study showed reciprocal linkages between physical activity and loneliness (Groß & Haffa, Reference Groß and Haffa2025). Together, these findings suggest that health behaviors may influence psychosocial functioning in complex and potentially bidirectional ways across the life course. Yet, no study has systematically investigated the dynamic interplay between loneliness and a broader range of health behaviors across the first half of life.
To address these evidence gaps, we applied a causal inference framework to data from a Norwegian population-based cohort spanning 28 years from adolescence to midlife, investigating prospective reciprocal associations between loneliness and key health behaviors, comprising alcohol use, BMI, smoking, and physical exercise. We hypothesized that loneliness would be prospectively associated with adverse health behaviors, and that these behaviors, in turn, would be linked to subsequent levels of loneliness.
Methods
Open science practices
The study was preregistered on the Open Science Framework (OSF) before conducting statistical analyses (OSF, 2025). To inform the study design, we inspected frequencies and distributions of loneliness and health behavior measures before preregistration. All deviations from the preregistered protocol are documented in the Supplementary Materials (p. 3).
Procedure and participants
This study is based on the population-based cohort study Young in Norway. Data were collected at five time points across 28 years: 1992 (T0, middle adolescence, mean age 15 years), 1994 (T1, late adolescence, 17 years), 1999 (T2, emerging adulthood, 22 years), 2005 (T3, young adulthood, 28 years), and 2020 (T4, middle adulthood, 43 years). At T0, a stratified, two-stage cluster probability sampling method was employed to draw a nationally representative sample of Norwegian adolescents aged mainly 13–18 years from 67 junior and senior high schools (n = 4,187; initial response rate [RR] = 97.0%). Participants were followed up at T1 (n = 3,844; RR = 91.8%), T2 (n = 2,923; RR = 83.8%), and T3 (n = 2,890; RR = 82.4%). Finally, at T4, 2,215 individuals participated (RR = 85.6%, cumulative RR = 52.9%). The present analytical sample encompassed 3,072 individuals responding to at least three data collection waves. A detailed participant flow chart is provided in the Supplementary Materials (Supplementary Figure S1).
To account for potential attrition bias and preserve representativeness of the sample, we performed all analyses using nonresponse weights using inverse probability weighting with a LASSO-selected set of baseline predictors (for details, see Supplementary Materials pp. 4–5). Missing data were handled with multiple imputations by chained equations using the R package mice (version 3.16.0). We generated 10 imputed datasets and pooled the estimates using Rubin’s rules.
All procedures adhered to the Helsinki Declaration, and we obtained informed consent from participants throughout the study. The Young in Norway Study was approved by the Norwegian Data Inspectorate and the Regional Committees for Medical and Health Research Ethics (reference 25462).
Measures
Loneliness
At all five waves, loneliness was assessed by a four-item short version of the revised UCLA Loneliness scale (Russell, Peplau, & Cutrona, Reference Russell, Peplau and Cutrona1980) (example item: ‘People are around me but not with me’). Responses were recorded on a four-point rating scale, ranging from ‘Never’ to ‘Often’. Composite mean scores were calculated and standardized (mean = 0, standard deviation [SD] = 1) to facilitate interpretation. McDonald’s omega (ω) ranged from 0.66 at T1 to 0.81 at T4, indicating satisfactory to very good scale reliability.
Health behaviors
Self-reported information on health behaviors was collected at each time point. Alcohol use was measured by multiplying drinking frequency over the past 4 weeks (‘How many times during the last 4 weeks have you been drinking more than a few sips of alcohol?’) by the number of drinks consumed on the last drinking occasion (‘Last time you drank alcohol, how many drinks [standard units] did you have?’), yielding an index of average 4-week alcohol use. Smoking behavior was assessed by one multiple-choice item (‘Do you smoke?’) and categorized as current smoking (daily [‘I smoke daily, about … cigarettes’] or occasional [‘I smoke, but not daily’]) versus nonsmoking (all other response options). BMI was calculated based on self-reported weight and height as weight (kg)/height (m)2. Physical exercise was measured from T2 to T4 by the item ‘How many hours did you spend on physical exercise last 7 days? (hours/minutes)’. At T0 and T1, participants reported 7-day frequency of three types of physical activity: ‘training with a sports team’, ‘exercise at a fitness center’, and individual physical activity (e.g. ‘jogging’ and ‘swimming’), and we calculated a sum of these activities. To increase consistency, physical exercise measures at T0 and T1 were transformed by multiplying the number of physical activities by 60 (minutes), rendering a harmonized indicator of physical exercise.
Confounders
To increase the confidence that the exchangeability assumption holds (i.e. no unmeasured confounding), we selected a comprehensive set of baseline and time-varying confounders based on previous studies on loneliness and health (Kim et al., Reference Kim, Wilkinson, Holt-Lunstad and VanderWeele2025; Kobayashi & Steptoe, Reference Kobayashi and Steptoe2018; Pengpid et al., Reference Pengpid, Peltzer and Anantanasuwong2023; Shankar et al., Reference Shankar, McMunn, Banks and Steptoe2011) and theoretical considerations. Baseline, time-invariant confounders (C 0, measured at T0) included sociodemographic characteristics (age, gender, migration background, and parental education), early family environment (parental warmth, overcontrol, and behavioral monitoring), parental health behaviors (parental alcohol use and smoking), and chronic health conditions (physical disability, asthma, and reading/writing disability). Time-varying confounders (Lt -1, measured from T0 to T3) comprised socioeconomic characteristics (employment status and educational attainment), indicators of social integration (living situation, relationship status, and number of friends), perceived social support, and depressive symptoms. For a directed acyclic graph showing the assumed causal structure, see Supplementary Figure S2, and detailed information on confounder measurement is provided in the Supplementary Materials (p. 6).
Statistical analysis
To examine reciprocal longitudinal associations between loneliness and health behaviors from adolescence to midlife, we applied structural mean modeling (SMM) (Mulder et al., Reference Mulder, Usami and Hamaker2025). This causal modeling approach, developed within the potential outcomes framework, is a propensity-score-based, doubly robust modeling approach that allows estimation of causal effects. By accounting for previous levels of loneliness and health behaviors, as well as a comprehensive set of baseline and time-varying confounders, SMM allowed us to minimize the possibility of reverse causation and residual confounding, and examine the effect of changes in exposure on subsequent changes in the outcome. Because links between health behaviors and psychosocial outcomes may be nonlinear, with adverse outcomes potentially emerging at both very high and very low levels of behavior, a U- or J-shaped relationship is plausible (Visontay et al., Reference Visontay, Mewton, Slade, Aris and Sunderland2023). Continuous health behaviors (alcohol use, BMI, and physical exercise) were therefore modeled using natural (restricted cubic) splines, a semi-parametric approach widely used in biomedical research for capturing complex, nonlinear dose–response relationships (Desquilbet & Mariotti, Reference Desquilbet and Mariotti2010; Ho & Cole, Reference Ho and Cole2023). A detailed description of the overall modeling strategy is provided in the Supplementary Materials (p. 7).
We note that while our analytical approach does not completely rule out the possibility of residual confounding, it greatly strengthens our confidence in causal inference. We considered a comprehensive set of confounders, which also serve as proxies for unmeasured confounders (e.g. indicators of early family environment and parenting as proxies for unobserved confounders such as genetic vulnerability), making the fulfillment of the exchangeability assumption more plausible. For a thorough discussion of the identifiability assumptions required for causal interpretation, including exchangeability, consistency, and correct model specification, see pp. 8–9 in the Supplementary Materials.
We examined prespecified exposure contrasts informed by the WHO (World Health Organization, 2020, 2024) and US (National Institute on Alcohol Abuse and Alcoholism, 2024) guidelines:
-
1. smoking (daily or occasional) versus nonsmoking (dichotomized);
-
2. adiposity (BMI = 30) and underweight (BMI = 17) versus normative BMI (BMI = 21.75) (continuous);
-
3. abstinence (0 units/4 weeks) and heavy alcohol use (60 units/4 weeks) versus moderate drinking (10 units/4 weeks) (continuous);
-
4. moderate physical exercise (150 minutes/week) versus no physical exercise (0 minutes/week) (continuous); and
-
5. high loneliness level (1 SD above the mean) versus the mean level (SD = 0) (continuous).
For each contrast, we calculated average effect estimates and 95% confidence intervals (CIs) based on robust (sandwich) standard errors. We additionally tested for moderation by gender and report gender-specific effect estimates.
All data preprocessing and statistical analyses were performed using the software R (version 4.2.3). We used the splines package (version 4.2.3) to model nonlinear dose–response relationships and the marginaleffects package (version 0.24.0) for computing effect estimates.
Sensitivity analyses
Several sensitivity analyses assessed the robustness of our findings. First, we reestimated models excluding social isolation, social support, and depressive symptoms from the set of confounders. Second, we assessed associations using heavy episodic drinking rather than total alcohol use. Third, we applied life stage-specific exposure contrasts when examining the effects of health behaviors. Specifically, given that the same level of health behaviors may reflect different substantive meanings across life stages (e.g. the number of alcoholic drinks in adolescence versus in young adulthood), we additionally specified SD-based exposure contrasts separately for each life stage (see Supplementary Table S4). Fourth, we replicated analyses using a direct single-item loneliness measure (‘I feel lonely’). Full rationale and technical details of the sensitivity analyses are provided in the Supplementary Materials (p. 10).
Results
At baseline (T0), the sample comprised 45.3% men and 54.7% women, with a mean age of 15.0 years (SD = 1.9). Most respondents had parents born in Norway (97.9%), while 2.1% had a migration background. Regarding parental educational attainment, 42.6% had parents with a higher education degree. A total of 3.0% had both parents currently unemployed. On average, participants engaged in 167.3 (SD = 157.0) minutes of physical exercise during the past week; consumed 7.4 (SD = 30.5) standard alcohol units in the past 4 weeks, and 19.4% were current smokers. The mean BMI was 20.0 (SD = 2.7), and the mean unstandardized loneliness score was 1.8 (SD = 0.5). Detailed sample characteristics at T0 are presented in Table 1. Descriptive statistics for loneliness and health behaviors from T1 to T4 are shown in Supplementary Tables S1 and S2 in the Supplementary Materials, and missing data statistics are displayed in Supplementary Table S3.
Sample characteristics at the baseline (T0, adolescence)
Table 1. Long description
The table presents data for a sample of adolescents across several categories.
Demographics and Background:
- Gender: Women 1679 (54.65 percent), Men 1393 (45.35 percent).
- Age: mean 14.95, S D 1.88.
- Migration background: Norway-born parents 2957 (97.88 percent), Migrant 64 (2.12 percent).
- Chronic health condition: 424 (13.80 percent).
- Parental education: College or university 1262 (42.62 percent), Lower level 1699 (57.38 percent).
- Parental employment: Employed 2924 (96.98 percent), In unemployment 91 (3.02 percent).
Parenting and Home Environment:
- Behavioral monitoring: mean 2.14, S D 0.89.
- Psychological overcontrol: mean 2.10, S D 0.56.
- Cold parenting: mean 1.83, S D 0.53.
- Parental alcohol use: Abstinence 337 (11.30 percent), Heavy use 393 (13.18 percent), Low use 2252 (75.52 percent).
- Parental smoking: Daily smoking 1319 (48.89 percent), Nonsmoking 1379 (51.11 percent).
- Living situation: Not with both parents 695 (23.04 percent), With both parents 2322 (76.96 percent).
Psychosocial and Health Behaviors:
- Depressive symptoms: mean 1.71, S D 0.56.
- Social support: mean 0.23, S D 0.50.
- Relationship status: In a relationship 602 (19.66 percent), Single 2460 (80.34 percent).
- Number of friends: A lot 2638 (87.93 percent).
- Alcohol use: mean 7.38, S D 30.53.
- Smoking status: Daily or occasional 572 (19.38 percent).
- Physical exercise: mean 2.79, S D 2.62.
- Body mass index: mean 20.00, S D 2.66.
- Loneliness: mean 1.84, S D 0.54.
Note: Statistics are based on the incomplete dataset before multiple imputations.
In a first step, we examined associations between loneliness and subsequent health behaviors across three life stages: late adolescence, emerging adulthood, and young adulthood. As displayed in Figure 1, a 1 SD increase in loneliness during late adolescence was prospectively associated with a 14-minute reduction in physical exercise (mean difference: −14.13 minutes/week, 95% CI: −26.54 to −1.72). Loneliness during young adulthood was associated with subsequently increased BMI (0.37 points, 95% CI: 0.12–0.63). Loneliness showed no significant effects on smoking and alcohol use, nor on physical exercise and BMI at all other life stages.
Estimates for causal effects of loneliness on health behaviors. Note: Labels of life stages below the panels refer to the exposure period.
Figure 1. Long description
A multi-panel figure containing four dot plots. Each plot shares a common x-axis with three life stages: adolescence (17y), emerging adulthood (22y), and young adulthood (28y). A horizontal dashed line at zero represents no effect.
1. Top-left panel: Alcohol Use. Y-axis is drinks per 4 weeks ranging from -3 to 1. Adolescence shows a point estimate near -0.5 with wide error bars from -2.3 to 1.5. Emerging adulthood is near -0.5 with bars from -2 to 0.9. Young adulthood is near 0.3 with bars from -1.1 to 1.8.
2. Top-right panel: Body Mass Index. Y-axis is k g / m super 2 ranging from -0.25 to 0.50. Adolescence is near 0.13 with bars from 0 to 0.27. Emerging adulthood is near -0.08 with bars from -0.25 to 0.1. Young adulthood is near 0.38 with bars from 0.12 to 0.63.
3. Bottom-left panel: Smoking. Y-axis is probability ranging from -0.03 to 0.03. Adolescence is near -0.007 with bars from -0.03 to 0.015. Emerging adulthood is near 0.002 with bars from -0.023 to 0.026. Young adulthood is near 0.008 with bars from -0.013 to 0.028.
4. Bottom-right panel: Physical Exercise. Y-axis is minutes per week ranging from -30 to 10. Adolescence is near -14 with bars from -26 to -2. Emerging adulthood is near 0 with bars from -12 to 12. Young adulthood is near -5 with bars from -16 to 6.
Next, we investigated the longitudinal associations between health behaviors and loneliness. As shown in Figure 2, alcohol abstinence during late adolescence and emerging adulthood, compared to moderate alcohol use, was linked to higher levels of subsequent loneliness by 0.07 (95% CI: 0.01–0.13) and 0.19 (95% CI: 0.05–0.33) SDs, respectively. Smoking in late adolescence was prospectively associated with diminished loneliness (−0.12 SD, 95% CI: −0.23 to −0.01). Similarly, moderate physical exercise during adolescence and emerging adulthood was linked to subsequently reduced loneliness levels (−0.16 SD, 95% CI: −0.26 to −0.07; and −0.07 SD, 95% CI: −0.14 to −0.01, respectively). Health behaviors in young adulthood were not associated with later loneliness.
Estimates for causal effects of health behaviors on loneliness. Note: Labels of life stages below the panels refer to the exposure period.
Figure 2. Long description
A multi-panel figure with four plots arranged in a two-by-two grid. The y-axis for all panels represents Loneliness standardized, ranging from negative 0.2 to 0.4. The x-axis for all panels lists three life stages: adolescence 17y, emerging adulthood 22y, and young adulthood 28y. A horizontal dashed line at 0.0 indicates the null effect.
Top-left panel: Alcohol Use. Reference is moderate. Yellow triangles represent abstinence, showing positive effects on loneliness that peak in emerging adulthood. Blue circles represent heavy use, showing slight negative effects across all stages.
Top-right panel: Body Mass Index. Reference is normative. Yellow triangles represent underweight, showing positive effects with wide error bars. Blue circles represent adiposity, showing a shift from positive in adolescence to negative in emerging adulthood, then positive again in young adulthood.
Bottom-left panel: Smoking. Reference is none. Blue circles represent smoking, showing a linear upward trend from a negative effect in adolescence to a positive effect in young adulthood.
Bottom-right panel: Physical Exercise. Reference is inactive. Blue circles represent moderate exercise, showing negative effects on loneliness across all three life stages, with the strongest negative effect in adolescence.
Moderation analyses by gender revealed distinct patterns. Among women, a 1 SD increase in loneliness during late adolescence was linked to a subsequently higher BMI (0.25 points, 95% CI: 0.07 to 0.43), and both underweight in emerging adulthood and adiposity in young adulthood were longitudinally linked to higher subsequent loneliness by 0.27 SD (95% CI: 0.02–0.51) and 0.24 SD (95% CI: 0.06–0.42), respectively. Among men, a 1 SD increase in loneliness during late adolescence was linked to lower levels of physical exercise (−26.29 minutes/week, 95% CI: −43.24 to −9.34), whereas moderate physical exercise in late adolescence was associated with reduced loneliness levels (−0.21 SD, 95% CI: −0.35 to −0.07). Detailed results are shown in Supplementary Figures S4 and S5 in the Supplementary Materials.
Results of sensitivity analyses were largely consistent with the main findings (Supplementary Figures S6–S11 in the Supplementary Materials). Reestimation of models using a reduced set of time-varying confounders revealed only marginal changes in the effect estimates. Notably, exclusion of theoretically relevant confounders, such as prior depressive symptoms, had only a negligible influence on the results, suggesting that our adjustment set was reasonably comprehensive. This pattern further suggests that the predominant null findings were not driven by specific modeling decisions or overadjustment and supports the robustness of the conclusions across alternative model specifications.
Given the substantial number of null findings for associations between loneliness and subsequent health behaviors, we performed post hoc positive control analyses to assess the validity of our analytical approach. Using the same modeling framework, we examined the well-established longitudinal associations between loneliness and depressive symptoms (Goosby et al., Reference Goosby, Bellatorre, Walsemann and Cheadle2013; Kim et al., Reference Kim, Wilkinson, Holt-Lunstad and VanderWeele2025). Results showed a consistent pattern of prospective associations between loneliness and depressive symptoms across all life stages (see Supplementary Figure S12 in the Supplementary Materials), supporting the robustness and sensitivity of our analytical approach to detect effects when present. Bivariate associations (i.e. unadjusted/crude estimates) between loneliness and subsequent health behaviors are shown in Supplementary Figure S13.
Discussion
In this Norwegian population-based cohort study spanning 28 years, we applied a causal inference approach to investigate the dynamic interplay between loneliness and health behaviors from adolescence to midlife. We showed that increased levels of loneliness in adolescence were longitudinally linked to lower physical exercise, and elevated loneliness in young adulthood was associated with subsequently increased BMI. All other associations between loneliness and subsequent health behaviors were consistent with null effects. In the reverse direction, moderate alcohol use, smoking, and moderate physical exercise in adolescence and emerging adulthood were prospectively linked to lower levels of loneliness. We also observed tentative gender-specific patterns: reciprocal loneliness–BMI associations in women, and bidirectional loneliness-physical exercise linkages among men.
Our findings, providing limited support for causal effects of loneliness on subsequent health behaviors, are broadly in line with a recent US cohort study employing a causal inference framework, reporting no prospective associations between adolescent loneliness and adult health behaviors, including smoking, alcohol use, physical activity, and overweight/obesity (Kim et al., Reference Kim, Wilkinson, Holt-Lunstad and VanderWeele2025). Similarly, these findings align with longitudinal studies demonstrating no loneliness-health behavior linkages in older adults (Hong et al., Reference Hong, Nakamura, Berkman, Chen, Shiba, Chen and VanderWeele2023; Kobayashi & Steptoe, Reference Kobayashi and Steptoe2018; Schrempft et al., Reference Schrempft, Jackowska, Hamer and Steptoe2019) and evidence showing weak genetic loneliness-obesity linkages (Liang et al., Reference Liang, Zhou, He, Zhang, Wu, Luo and Zhang2024). However, our findings contrast with longitudinal studies showing that loneliness is longitudinally associated with physical inactivity (Hawkley et al., Reference Hawkley, Thisted and Cacioppo2009; Pengpid et al., Reference Pengpid, Peltzer and Anantanasuwong2023; Shankar et al., Reference Shankar, McMunn, Banks and Steptoe2011) and smoking (Shankar et al., Reference Shankar, McMunn, Banks and Steptoe2011) among older adults. Two complementary explanations may account for these discrepancies. First, unlike most previous cohort studies, we applied a causal inference approach that adjusted for prior levels of loneliness and health behaviors, as well as a comprehensive set of baseline and time-varying confounders. This approach enabled estimation of the effects of incident increases in loneliness on subsequent changes in health behaviors, thereby reducing the risk of residual confounding and reverse causation, leading to less biased estimates than in conventional approaches. Second, whereas previous studies largely focused on older adults (Hong et al., Reference Hong, Nakamura, Berkman, Chen, Shiba, Chen and VanderWeele2023; Kobayashi & Steptoe, Reference Kobayashi and Steptoe2018; Pengpid et al., Reference Pengpid, Peltzer and Anantanasuwong2023; Shankar et al., Reference Shankar, McMunn, Banks and Steptoe2011), our study examined the first half of life, a period when smoking, alcohol use, and physical exercise are more socially embedded. During adolescence and emerging adulthood, health behaviors may be shaped primarily by peer and social contexts (Montgomery et al., Reference Montgomery, Donnelly, Bhatnagar, Carlin, Kee and Hunter2020) rather than by internal psychological states such as loneliness. By contrast, in older age, when solitary smoking and alcohol use tend to be more commonly endorsed and socially accepted, loneliness may have a more pronounced effect on these adverse health behaviors.
Given the limited support for the causal effects of loneliness on health behaviors, other pathways may underlie the links between loneliness and subsequent physical health and mortality. In our positive control outcome analysis, elevated loneliness was consistently associated with increases in depressive symptoms across all examined life stages, aligning with prior evidence demonstrating robust links between loneliness and subsequent depressive symptomatology (Goosby et al., Reference Goosby, Bellatorre, Walsemann and Cheadle2013; Kim et al., Reference Kim, Wilkinson, Holt-Lunstad and VanderWeele2025). Importantly, depression and other mental health problems are linked to increased risk of a wide range of adverse physical health conditions (Momen et al., Reference Momen, Plana-Ripoll, Agerbo, Benros, Børglum, Christensen and McGrath2020), and these conditions are in turn related to higher mortality among individuals with preexisting mental health conditions (Formánek et al., Reference Formánek, Krupchanka, Mladá, Winkler and Jones2022, Reference Formánek, Krupchanka, Perry, Mladá, Osimo, Masopust and Plana-Ripoll2024). One plausible interpretation is therefore that loneliness in young people may primarily contribute to poorer mental health, which, in turn, may increase vulnerability to adverse physical health outcomes and premature mortality later in life.
When examining the effect of health behaviors on loneliness, we demonstrated that moderate physical exercise, moderate alcohol use, and smoking during adolescence and emerging adulthood were related to lower levels of loneliness. These findings suggest that health behaviors in young people – whether health-promoting or risky – may contribute to reducing loneliness. This pattern contrasts with a recent cohort study in England, showing that smoking among older adults was associated with increased loneliness (Philip et al., Reference Philip, Bu, Polkey, Brown, Steptoe, Hopkinson and Fancourt2022). However, our results align with a US cohort study employing a causal inference approach, which demonstrated that moderate alcohol use in young adulthood was associated with a reduced risk of depression (Visontay et al., Reference Visontay, Mewton, Slade, Aris and Sunderland2023) and an Australian household panel study showing longitudinal links between physical activity and reduced loneliness (Groß & Haffa, Reference Groß and Haffa2025). One potential explanation for these conflicting findings relates again to the social embeddedness of health behaviors. During adolescence and emerging adulthood, behaviors such as smoking, alcohol use, and physical exercise often occur in peer contexts, where they may facilitate socializing, provide social validation, and strengthen a sense of belonging (Montgomery et al., Reference Montgomery, Donnelly, Bhatnagar, Carlin, Kee and Hunter2020), thereby reducing loneliness. These mechanisms may be less relevant at later life stages. Period and cohort effects may also help to explain the results. Our participants were adolescents in the 1990s, when smoking and alcohol use were common and socially accepted in Norway. In this socio-historical context, engaging in these behaviors may have aligned with prevailing norms and fostered social connectedness. By contrast, as some of these behaviors have become less common and increasingly stigmatized, they may now contribute to social marginalization (Christakis & Fowler, Reference Christakis and Fowler2008; von Soest & Pedersen, Reference von Soest and Pedersen2014) and heightened loneliness (Philip et al., Reference Philip, Bu, Polkey, Brown, Steptoe, Hopkinson and Fancourt2022).
Gender-stratified analyses showed that elevated loneliness during adolescence was related to increased BMI, whereas both underweight in emerging adulthood and adiposity in young adulthood were longitudinally associated with increased loneliness among women. These findings are consistent with longitudinal evidence suggesting an effect of loneliness on BMI (Goosby et al., Reference Goosby, Bellatorre, Walsemann and Cheadle2013) and Mendelian randomization evidence indicating effects of BMI on loneliness (Abdellaoui et al., Reference Abdellaoui, Sanchez-Roige, Sealock, Treur, Dennis, Fontanillas and Boomsma2019). Given pervasive norms around the female body, women with nonnormative BMI may be more likely than men to experience body dissatisfaction and feelings of shame and guilt, potentially leading to social withdrawal or rejection by others. Such mechanisms may be particularly salient during emerging and young adulthood, when key social and romantic relationships are established. Among adolescent men, consistent with prior evidence (Groß & Haffa, Reference Groß and Haffa2025), we showed that loneliness was prospectively associated with decreased levels of physical exercise, whereas moderate physical exercise was linked to lowered loneliness. This suggests that sports and physical activity may serve as important and socially valued arenas for male adolescent socialization.
The present study has several strengths, including the use of nationally representative cohort data spanning nearly three decades from adolescence to midlife, as well as the application of a causal inference approach designed to minimize residual confounding and reverse causation. However, several limitations should be acknowledged. First, the relatively long time intervals between data collection waves might have been suboptimal for capturing short-term causal dynamics (Groß & Haffa, Reference Groß and Haffa2025). Longitudinal designs with more frequent assessments may better reflect the temporal fluctuations in loneliness and health behaviors. Second, future research should consider the social context in which health behaviors occur – specifically, whether they are socially or solitarily enacted – as this may moderate their associations with loneliness. Third, while our study examined situational changes in loneliness, more persistent feelings of social disconnection may have stronger and more enduring effects on health. Investigating the long-term impact of chronic loneliness across the first half of life is therefore warranted. Fourth, alcohol use was not legal for most respondents at the first examined life stage (in adolescence, mean age = 17 years). Although we cannot rule out underreporting bias, the pattern of associations with subsequent loneliness at this life stage was largely similar to that in emerging adulthood (mean age = 22 years), when alcohol consumption is legal. This consistency suggests that reporting bias is unlikely to fully explain the observed findings. Fifth, this study was conducted in Norway, an egalitarian welfare state with a high standard of living, which may limit the generalizability to non-Western or less affluent countries. Finally, the single-cohort design precluded disentangling age (e.g. adolescence), period (e.g. the 1990s), and cohort (e.g. Millennials) effects. Future research should therefore examine whether associations vary across birth cohorts, geographical and historical contexts, and examine how shifting social norms influence the relationship between loneliness and health behaviors.
Conclusion
In this Norwegian population-based cohort study spanning 28 years from adolescence to midlife, we applied a causal inference approach to investigate the dynamic interplay between loneliness and health behaviors. We found little evidence supporting causal pathways from loneliness to subsequent adverse health behaviors across the first half of life. In the opposite direction, moderate alcohol use, smoking, and moderate physical exercise in adolescence and emerging adulthood were prospectively linked to reduced loneliness levels. These findings suggest that health behaviors – whether health-promoting or risky – are socially embedded and may serve as important avenues for socializing in young people, underscoring the importance of a life course approach to interventions targeting social well-being. To address youth loneliness, public health strategies should promote equitable access to health-promoting behaviors that foster social inclusion and belonging. This may include, for example, institutional support for organized sports activities, ensuring that young people have access to inclusive sports environments.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726105029.
Data availability statement
Due to the sensitive nature of the data and data protection requirements, the data are not publicly available. The complete analytical R script is available here: https://github.com/libpot/Loneliness_Health_Factors.
Acknowledgments
This work was supported by the Research Council of Norway (grant numbers 288083, 301010, and 300816). We are enormously grateful to all the people who have participated in the Young in Norway study throughout the decades.
Author contribution
L.P. initiated, conceptualized, and designed the study, performed the literature search, methodology, investigation, formal analysis, visualization, and led the writing of the manuscript. L.C.A. contributed to the study’s conceptualization, design, methodology, and interpretation of results and provided critical revisions to the manuscript. M.K. contributed to study design, methodology, and interpretation of results, conducted data validation, and provided supervision and critical revisions to the manuscript. R.R.C. contributed to the study design and methodology and provided critical revisions to the manuscript. T.F. conceptualized and designed the study and contributed to the methodology, interpretation of results, and provided supervision and critical revisions to the manuscript. T.v.S. secured funding, curated the data, conceptualized and designed the study, contributed to the methodology and interpretation of results, and provided supervision and critical revisions. L.P., M.K., and T.v.S. had full access to all data. T.v.S. takes responsibility for the integrity of the data, and L.P. for the accuracy of the data analysis. All authors approved the decision to submit for publication.
Funding statement
This work was supported by the Research Council of Norway (grant numbers 288083, 301010, and 300816). The funder of the study had no role in study design, data collection, data analysis, data interpretation, writing of the report, or in the decision to submit the paper for publication.
Competing interests
The authors declare no competing interests.
Open access
