Across three countries, couples keep matching on mental health. In a Nature Human Behaviour study analyzing nearly 5 million couples in Taiwan and comparing with 1.3 million in Denmark and Sweden, researchers report consistent spousal correlations across nine psychiatric disorders.
Using national registries, the team shows that partners with conditions like major depression, schizophrenia, bipolar disorder and substance use disorder are more likely to pair with someone who shares a diagnosis or related risk, and that these patterns have held from cohorts born in the 1930s through the 1990s.
What it means when partners resemble each other
Spousal correlation sounds technical, but the idea is simple. People with similar traits often end up together. Sometimes that is assortative mating, an active preference for similarity. Sometimes it is convergence, life lived together pulling partners closer. Sometimes it is social homogamy, the quiet pressure of schools, neighborhoods and norms. In psychiatric disorders, the stakes rise. If similarity is consistent, not a passing quirk of place or policy, it can nudge prevalence, complicate comorbidities, and bend the genetic signals we think we are reading.
Across nations, a shared pattern of spousal correlations
The new analysis begins in Taiwan’s National Health Insurance Research Database, which covers 99 percent of the population. From nearly five million legally recognized spousal pairs, the authors estimate tetrachoric correlations for nine disorders, then compare them to matched case–control estimates from Denmark, and published estimates from Sweden. The headline result is repetition by design: similarity is widespread. Most disorder pairs show significantly positive spousal correlations, generally aligning with Nordic estimates. Differences emerge for anorexia nervosa, obsessive compulsive disorder and bipolar disorder, but the broad story holds, country to country.
Across nine psychiatric disorders, there is evidence of positive correlations among spouses, independent of country and generation, for the past 90 years.
Numbers ground the scene. The study includes 1.4 million psychiatric cases among probands and up to six million matched controls. Correlations span a range roughly from 0.05 to 0.57. For substance use disorder, multiple cross-trait pairings trend upward across decades. For obsessive compulsive disorder, the opposite emerges, a downward trend that echoes a Swedish observation. A small aside, perhaps impolitic, but here it is: when two different national registries hum the same tune, you lean in.
Generations do change, but the pattern endures
To test whether modern life has softened these effects, the authors stratify Taiwanese couples by birth decade, from the 1930s to the 1990s. Because tetrachoric correlations are relatively insensitive to prevalence shifts, they offer a clean read on change. Sixteen of 81 disorder pairs show study-wide significant generational trends. Most are increases, especially those involving substance use disorder. OCD, by contrast, declines. ADHD shows some large swings, but the overall trend is not significant after strict correction. The authors consider diagnostic censoring and show that it would, if anything, dampen correlations, not inflate them.
From couples to children, the signal carries forward
If partner similarity is driven by assortative mating, theory predicts a downstream rise in parent–offspring resemblance. The data comply. In Taiwan, correlations between parents and children increase over generations for major depression and substance use disorder, and decline for OCD, closely tracking the spousal trends. Risk multiplies when both parents share a diagnosis. For schizophrenia, the odds ratio for offspring exceeds four when both parents are affected. It is a sober detail, but clarity helps families and clinicians evaluate risk honestly.
This observed phenomenon may contribute to the co-occurrence of psychiatric disorders and bias genetic estimates of two biologically unrelated but assorted disorders.
Spousal correlations and the genetic mirror
There is a second mirror here. The authors compare Taiwan’s spousal correlations with GWAS-based heritability and genetic correlations derived largely from European cohorts. Despite demographic differences, the alignment is statistically significant. Around one fifth of the variance in observed spousal correlations is explained by GWAS-based genetic covariance. The point is not that one causes the other, rather that persistent non-random mating and genetics meet in the middle, and that standard models assuming random mating may need revision. How often do our clean assumptions blur the picture we think we see?
Limits, nuance, and what to do next
These are observational registries, not randomized experiments. Diagnoses can arrive after marriage. Convergence and clinical practices could add noise. Nordic generational data were not available for a like-for-like trend analysis. Even so, the cross-country consistency, the decade-by-decade coherence, and the parent–offspring correspondence tell a steady story. For researchers, that means treating non-random mating not as a curiosity but as a design consideration. For clinicians and families, it means acknowledging that partner similarity in mental health is common and durable, and that risk can concentrate when two similar histories meet.
Repetition can be a flaw in prose, but in population science it is a strength. Across registries, across decades, the pattern repeats.
Explainer: What are “spousal correlations” and why use tetrachoric r?
Spousal correlation is the statistical similarity between partners for a trait, here psychiatric diagnoses. Because diagnoses are yes/no variables, the study uses a tetrachoric correlation, which assumes an underlying continuous liability and estimates the correlation that would generate the observed binary outcomes. That approach is less sensitive to shifts in prevalence across time or place, which is crucial when comparing birth cohorts from the 1930s through the 1990s or matching Taiwan’s registry to Nordic registries. The study also uses a matched case–control design, pairing each diagnosed proband with five controls similar in sex, age, calendar year and region. Together, these choices reduce confounding and allow cleaner comparisons of cultural and generational trends.
Journal: Nature Human Behaviour (2025)
ScienceBlog.com has no paywalls, no sponsored content, and no agenda beyond getting the science right. Every story here is written to inform, not to impress an advertiser or push a point of view.
Good science journalism takes time — reading the papers, checking the claims, finding researchers who can put findings in context. We do that work because we think it matters.
If you find this site useful, consider supporting it with a donation. Even a few dollars a month helps keep the coverage independent and free for everyone.