Relationships and Happiness

Connection and long life

In 2010 Julianne Holt-Lunstad and colleagues pooled 148 studies of 308,849 people. Those with stronger social relationships had about 50% higher odds of surviving over the study periods. The link was strongest for rich measures of social life (how many roles, groups and ties people have) and weakest, and not clearly reliable, for simply living alone. A 2015 follow-up found loneliness, social isolation and living alone were each linked to about 26% to 32% higher odds of earlier death.

Newer and larger reviews confirm the pattern but trim the loneliness figure. A 2023 review of 90 studies and over 2.2 million people found a 32% higher risk of death for social isolation and 14% for loneliness. A 2025 review of older adults found almost identical numbers. The logic matters here: isolation (few contacts) and loneliness (feeling alone) are different, and isolation seems more closely tied to physical health and death, while loneliness is more closely tied to mood.

All of these are observational studies. People who are ill may become isolated, rather than the other way round. Researchers adjust for this, but cannot remove it completely.

Quality, not just quantity

A review of 126 studies (over 72,000 people) found that happier marriages go with better health, but the effects were small, about the same size as the links between diet and health. In older adults, stronger friendships predicted lower risk of death, more exercise and better mood, but also more smoking, perhaps because friends tend to share habits, good and bad.

The Harvard Study of Adult Development

This study has followed the same American men since 1938, and later their wives and children. Its director summarises it as showing that a good life is built on relationships rather than achievement or riches. The papers we read support parts of this. In couples in their 80s, people were happier on days they spent more time with others, and a happy marriage buffered their mood on days when they felt less healthy. Feeling securely attached to a spouse predicted less depression two and a half years later. But the original samples were all male and white, and the findings are correlational.

What helps with loneliness?

A 2011 review by Christopher Masi and colleagues found four main approaches: social skills, social support, more chances to meet people, and changing unhelpful thoughts about others. Better-designed trials found smaller effects, and the most successful approach targeted unhelpful thinking, such as expecting rejection.

Recent work agrees. Talking-therapy approaches reduce loneliness by a moderate amount, though trials are small and at risk of bias. A 2026 trial in Hong Kong with 1,151 lonely older adults found that eight short phone sessions of either planning enjoyable activities or mindfulness reduced loneliness for a year, compared with friendly chat calls. The difference was small. For young people, the best trials show only a small effect that could be chance. Simply arranging more contact helps least. Technology on its own shows no reliable effect, although some online group therapy programmes do help.

AI companions

In a four-week trial of 981 adults, people who chose to chat with an AI chatbot more each day tended to be lonelier and to socialise less with real people. This was a correlation, the study had not been peer reviewed when we read it, and four of its authors work at OpenAI, which makes the chatbot.

If low mood is severe, lasts for weeks, or comes with thoughts of self-harm, talk to a doctor or a trusted adult.

What this means for you

The link between relationships and wellbeing is one of the most consistent in this field, even if exact cause and effect is uncertain. The most useful lever may be your own expectations: people reliably expect conversations to go worse than they do. Small, regular acts of contact are realistic and cheap, and there is little downside.

Beyond people: nature and animals

Relationships are the strongest everyday link to wellbeing. Researchers have asked whether connections with nature and animals add anything. The short answer: nature connection has a small, consistent link with wellbeing; pets have no reliable effect on average; and the treatment evidence for nature and animals is promising but mostly weak.

Nature connectedness. A 2014 meta-analysis of 30 samples (8,523 people) found that people who feel more connected to nature report more positive feelings, vitality and life satisfaction (Capaldi et al., 2014). A follow-up review found a similar link with a deeper kind of wellbeing, especially a sense of personal growth (Pritchard et al., 2020). Both effects are small, and almost all of the studies are surveys, so they cannot show that nature causes the happiness.

Nature with people. One of the clearest 2026 results joins this page's two themes. In a Finnish randomised trial, 319 lonely older adults (average age 83) either joined nine weekly small-group nature outings or received usual care. Loneliness fell in the nature groups at three months but not in the comparison group, though the difference faded by six to twelve months (Pitkälä et al., 2026). Since the groups were social, we cannot tell how much was the nature and how much the company.

Green space and young people. A Danish register study of about a million people found that those with the least green space around their childhood home had up to 55% higher risk of a range of later psychiatric disorders, even after adjusting for wealth and family history (Engemann et al., 2019). Reviews of adolescent studies mostly find similar favourable links (Zhang et al., 2020; Sprague et al., 2022). But in a 2025 analysis of 3,595 children followed from age 3 to 17 in England, green space was not directly linked to mental health once neighbourhood wealth was considered; wealth was the strongest factor (Shoari et al., 2025). Our reading: green space probably helps a little, but it is tangled up with income, safety and other neighbourhood features.

Forest bathing. Recent reviews pool 11 to 25 studies and report medium to large short-term drops in stress, anxiety and low mood after forest sessions (Qin et al., 2026). But results vary hugely between studies, participants cannot be blinded, many studies are small, and one 2026 review rated the overall certainty of the evidence as very low (Andrade et al., 2026). Treat forest bathing as a pleasant, low-risk practice, not a treatment.

Pets. A 2025 meta-analysis of 21 observational studies (159,322 people) found pet owners were no more and no less likely to be depressed than non-owners (Moshfeghinia et al., 2025). In 9,631 US teenagers, owning a pet did not predict later depressive symptoms (Żebrowska et al., 2022). In Hungary, people who got a pet during the pandemic felt more cheerful for one to four months, but over six months new dog owners reported lower calm and life satisfaction (Mokos et al., 2025). The main problem with pet research is confounding: people who choose and can afford pets differ from people who do not.

Animal-assisted interventions and horses. A 2026 meta-analysis of 35 randomised trials (2,391 adults) found that structured sessions with animals reduced depression by a small to moderate amount and anxiety and stress by more, though results varied widely (Shih et al., 2026). For veterans with PTSD, a 2024 review of 13 equine studies found lower PTSD scores after horse programmes, but only one study was at low risk of bias (Provan et al., 2024). The best known US studies are a small randomised trial of therapeutic riding (29 veterans; Johnson et al., 2018) and an uncontrolled Columbia University trial in which about half of 63 veterans improved markedly (Fisher et al., 2021). We could not find a published randomised trial from the Columbia group.

Time in nature and with animals is worth having, especially shared with other people. Expect a modest, mostly short-term lift. A pet is a relationship with costs as well as rewards, not a wellbeing treatment.

Mortality meta-analyses

Holt-Lunstad et al. (2010): k = 148, N = 308,849, OR = 1.50 (95% CI 1.42 to 1.59) for survival with stronger relationships; complex social integration OR = 1.91 (1.63 to 2.23); living alone OR = 1.19 (0.99 to 1.44). Holt-Lunstad et al. (2015; adjusted estimates): isolation OR = 1.29, loneliness 1.26, living alone 1.32, stronger in samples under 65. Wang et al. (2023; 90 cohorts, N = 2,205,199, PROSPERO-registered): all-cause mortality 1.32 (1.26 to 1.39) for isolation and 1.14 (1.08 to 1.20) for loneliness. Nakou et al. (2025; 86 studies, older adults): loneliness HR 1.14 (1.10 to 1.18, I² = 84%), isolation 1.35, living alone 1.21. Estimates converge on about 1.3 for isolation and 1.1 for loneliness, but residual confounding by health, depression and income is likely. Hong et al. (2023; N = 13,752), adjusting for prior exposures and outcomes, found isolation more predictive of mortality and loneliness of psychological outcomes.

The popular "15 cigarettes a day" comparison comes from the 2023 US Surgeon General's advisory; the meta-analyses themselves say only that effects are comparable to established risk factors. The WHO Commission (2025) attributes about 871,000 deaths a year to loneliness; the pages we read do not explain how this was modelled.

Mechanisms

Shen et al. (2025; UK Biobank, N = 42,062, 2,920 plasma proteins) linked isolation and loneliness to inflammatory, antiviral and complement proteins. Mendelian randomisation suggested a causal path from loneliness to five proteins, which partly mediated links with cardiovascular disease, stroke and mortality over 14 years: a mechanism hypothesis, given MR's strong assumptions. Robles et al. (2014) found marital quality associated with health at r = .07 to .21 (mortality r = .11).

Everyday interaction and wellbeing

Sun et al. (2020; N = 256, 3,206 observations, audio plus experience sampling) found interaction quantity robustly associated with wellbeing by self- and observer-report; quality effects were larger for self-report, suggesting some shared-method inflation. Sandstrom and Dunn (2014) linked more weak-tie interactions on a given day to greater happiness and belonging.

A programme of experiments from Epley's group shows systematic pessimism: people overestimate the awkwardness of deep talk with strangers (Kardas et al., 2022), underestimate how positively recipients respond to support (Dungan et al., 2022), and lose corrected expectations within one to two weeks (Atir & Epley, 2026). These are short-term US studies.

Friendship, contagion and the dark side

Kim et al. (2023; N = 12,998) linked stronger friendship to lower mortality and better psychosocial outcomes, but also to more smoking. In adolescents, co-rumination prospectively predicted depressive symptoms (46 datasets, N = 18,222, β = 0.036), a small effect driven by co-brooding (Li et al., 2026). Fowler and Christakis (2008; N = 4,739) reported that a nearby friend becoming happy raised one's own probability of happiness by 25% (95% CI 1% to 57%). In the same issue, Cohen-Cole and Fletcher (2008) showed similar methods detect "contagion" of acne and height that vanishes after adjusting for shared environment.

Development and the Harvard Study

Beyond the couple studies in Level 2, Nevarez et al. (2017; 101 WWII combat veterans) found good relationships with fellow soldiers weakened the link between combat exposure and PTSD symptoms. Limits: all-male, white US core samples, small subsamples, correlational designs. Funding: two of the papers we read list US National Institutes of Health support; we found no John Templeton Foundation statement in them and cannot confirm such funding. The best-known popular summaries are written by the Study's own directors.

Kim et al. (2025; Add Health, N = 11,040 and 9,003) found rising adolescent loneliness associated with 4 of 7 adult mental health outcomes (depression RR 1.25, 95% CI 1.06 to 1.49) but 1 of 13 physical outcomes. Funding: Kim, Wilkinson and VanderWeele (Kim et al., 2025), and Chen as well on Kim et al. (2023) and Hong et al. (2023), list affiliations with Harvard's Human Flourishing Program, whose flagship Global Flourishing Study is funded by the John Templeton Foundation, other Templeton charities and further foundations (Baylor University, 2021). Study-specific funding was not stated in the abstracts we read.

Interventions

Masi et al. (2011) found randomised designs gave smaller effects than pre-post designs, with maladaptive social cognition the most effective target. Birken et al. (2026; 22 studies, N = 2,567, all at some or high risk of bias) estimated g = -0.46 for CBT-based approaches. Coppola et al. (2025) reported SMD = -1.11 in older adults, but with I² = 99.6%. Across 40 digital RCTs (N = 6,062), group-based psychological interventions worked, while social contact interventions, conversational robots and social media reduction showed little (Hansen et al., 2025); technology-based interventions overall gave SMD = -0.21 (95% CI -0.59 to 0.17; Meier et al., 2026). In youth, six RCTs gave g = -0.20 (95% CI -0.42 to 0.02), with social-emotional skills training most promising (Burke et al., 2026). The strongest single trial is HEAL-HOA (Tang et al., 2026; N = 1,151, assessor-blinded, active befriending control): telephone behavioural activation and mindfulness reduced UCLA loneliness at 12 months by about 0.7 points, a small absolute difference.

AI and loneliness

Fang et al. (2025; preprint, N = 981, 4 weeks) found no effect of chatbot condition on loneliness, but heavier voluntary use correlated with more loneliness and less socialising. Conflict of interest: four authors are OpenAI employees. Cachia et al. (2026; N = 486, 6 weeks, care-as-usual control) found a generative-AI wellbeing app reduced loneliness but not depression or anxiety. Conflict of interest: the first two authors work for the app's developer, Flourish Science.

Claims we deliberately do not make

That loneliness is "as deadly as 15 cigarettes a day" (the meta-analyses do not support that precise equivalence); that relationships are proven to cause longer life (the evidence is observational); that the Harvard Study proves relationships matter more than anything else (its core samples were small, all-male and white); that happiness is contagious (the method critique leaves this unproven); that one close friend is enough and network size does not matter, as our previous page said (interaction quantity and weak ties also predict wellbeing); that apps, chatbots or social media breaks reliably reduce loneliness; and that any loneliness programme is proven for teenagers.

Beyond people: nature and animals

Nature connectedness (correlational). Capaldi et al. (2014): fixed-effect meta-analysis, 30 samples, N = 8,523, r = .19 with happiness (vitality r = .24, positive affect r = .22, life satisfaction r = .17); inclusion of nature in self showed the largest association (r = .27). Pritchard et al. (2020): 20 samples, n = 4,758, r = .24 with eudaimonic wellbeing, not significantly different from hedonic wellbeing (r = .20; 30 samples, n = 11,638); personal growth showed a moderate effect. Both syntheses are cross-sectional and self-report, so shared method variance and reverse causation cannot be excluded. The Pritchard group (Derby) runs a nature connectedness research programme; we found no commercial conflict.

Experimental nature exposure. A meta-analysis of RCTs of green exposure reported reduced fatigue (SMD -0.84), raised vitality (0.85) and lower heart rate (-0.60), and the authors called the effects large (Song et al., 2022). These are short, acute exposures, and the samples underrepresent adolescents. The Friends in Nature RCT (RECETAS, EU Horizon 2020; participants from 22 assisted-living facilities, n = 162 vs 157 usual care; 55% with dementia) found a between-group loneliness difference at 3 months (mDJGLS change -2.2 vs -0.1; p = .025) that attenuated by 6 and 12 months, no HRQoL effect, and better 15D sleep scores (Pitkälä et al., 2026). Secondary analyses reported small 3-month gains in clock drawing and in the Nature Connection Index relative to controls (Pitkala et al., 2026, JAMDA). Nature and group contact are confounded by design. Participants knew which group they were in, and loneliness was self-reported, which is a further limitation.

Green space and youth mental health. Engemann et al. (2019): Danish registers, about 1 million people, NDVI within 210 x 210 m from birth to age 10; lowest versus highest green space associated with up to 55% higher risk of later disorders, robust to urbanisation, socioeconomic status and parental history, with a cumulative-exposure gradient. Sprague et al. (2022): 28 prospective studies; 79% suggested benefit, concentrated in wealthy Western European settings. Zhang et al. (2020): 14 studies, favourable associations with stress, mood and distress but limited and at risk of bias. Bray et al. (2022): 48 studies in 14 to 24 year olds; experiments show immediate mood and state anxiety gains, with social interaction, physical activity and mindfulness proposed as mediators. Recent cohorts are mixed. In the Millennium Cohort Study, Shoari et al. (2025; n = 3,595) found neighbourhood socioeconomic status was the dominant correlate of SDQ scores, with no direct green space association (only a sex interaction). Baranyi et al. (2026; n ≈ 8,000) found a composite of built and social environment (lower greenness, more grey space, deprivation, crime) associated with high distress at 17 (OR about 1.2), with greenness not isolable from deprivation. In a GEMA study of 122 French adolescents, social company was associated with momentary wellbeing (being with friends B = 3.5), while green space showed only a trend (Buzzi et al., 2026). Residual confounding by socioeconomic position is the central threat to causal inference throughout.

Shinrin-yoku. Yeon et al. (2021): 20 studies, mostly Korean, depression g = 1.13 and anxiety g = 1.72, with 5 RCTs all of some concern or high risk of bias. One author was affiliated with Korea's National Center for Forest Therapy and one with the Korea Forest Therapy Forum, and most of the others with a university forest therapy department, so the team has an institutional interest in the practice. Qin et al. (2026): 25 RCT and quasi-experimental studies, N = 1,876; stress SMD -0.71, depression -0.68, anxiety -0.77; I² > 74%; funnel asymmetry for emotional outcomes, attenuated but still significant after trim-and-fill; GRADE moderate (stress) to low. Lu et al. (2026): 18 RCTs, N = 2,169, SMD -1.01 for depressive symptoms, I² = 90%, publication bias with 7 imputed studies. Andrade et al. (2026): 11 studies, GRADE very low, recommending it as a complement to health promotion only. Effect sizes this large in unblinded, self-report, short-follow-up trials should be read as upper bounds.

Pet ownership. Moshfeghinia et al. (2025): 21 observational studies, N = 159,322, pooled OR 1.03 (0.995 to 1.07); cats OR 1.06 (1.02 to 1.09); dogs OR 0.93 (0.79 to 1.10). Żebrowska et al. (2022; GUTS, n = 9,631, ages 12 to 18, propensity-adjusted): OR 1.14 (0.95 to 1.38), with effect modification by maternal depression. Endo et al. (2020; Tokyo Teen Cohort, n = 2,584) found dog ownership at age 10 predicted maintained wellbeing at 12 and cat ownership predicted decline. Mueller et al. (2021) found pet ownership predicted greater loneliness increases in adolescents during COVID-19. Northrope et al. (2025) reviewed 116 mostly cross-sectional studies of pet attachment: 15 favourable, 22 unfavourable, 36 mixed, 33 null; stronger attachment tracked worse symptom measures but better wellbeing measures. Mokos et al. (2025) used a stratified random Hungarian sample (n = 2,783; 65 acquirers), reducing the pet-enthusiast selection bias of earlier work: acquisition raised cheerfulness for 1 to 4 months, followed by declines in calm, activity and life satisfaction among dog acquirers by 6 months. Funding note: parts of this field are funded by the pet industry; for example, the Purewal et al. (2017) review of child and adolescent development was funded by the WALTHAM Centre for Pet Nutrition (Mars Petcare), which also employed one of its authors.

Animal-assisted interventions (AAI). Shih et al. (2026): 35 RCTs, N = 2,391 adults; depression g = -0.40, anxiety g = -0.66, stress g = -1.06; substantial heterogeneity; no effect on pain or gait. Lv et al. (2026): 20 studies in older adults; depression SMD -0.45 (GRADE low), no pooled effect on anxiety or loneliness, larger effects in quasi-experimental designs (a sign of design-related inflation). Control conditions rarely match the attention and novelty of an animal visit.

Equine-assisted services for veterans with PTSD. Johnson et al. (2018): randomised waitlist design, n = 29 (15 vs 14), 6 weeks of therapeutic riding; the abstract reports significant within-group PCL-M reductions at 3 and 6 weeks but no significant effects on coping self-efficacy, emotion regulation or loneliness, and emotional loneliness moved in the unexpected direction. Fisher et al. (2021): manualised 8-session group EAT-PTSD, open trial, n = 63, CAPS-5 38.6 to 26.9, 50.8% with ≥ 30% reduction post-treatment and 54.0% at 3 months, 8% dropout; no control group. Provan et al. (2024): 13 studies, 344 participants, PCL mean difference 12.46 (9.03 to 15.88), only one study at low risk of bias. Kowalski et al. (2025): retrospective multi-site pre-post evaluation (n = 107, 12 sites) of the Eagala model, co-authored by a representative of the Eagala association, which developed the model. We searched PubMed and ClinicalTrials.gov and found no published 2024 Columbia randomised trial; the Columbia (Man O' War Project) evidence is the 2021 open trial, and an RCT is listed as a goal of the project. We could not confirm the project's funding sources.

Claims we deliberately do not make. That nature or pets "cause" happiness; that green space prevents mental illness (the evidence is observational and confounded by wealth); that forest bathing works through phytoncides or immune changes, which we did not assess; that getting a pet will lift mood or protect a teenager from depression; that cat ownership causes depression; that horse therapy is an established PTSD treatment comparable to trauma-focused psychotherapy; that a randomised trial has shown equine therapy works for veterans.