A large new study analysing genetic data from the United Kingdom and the United States has found detectable footprints of assortative mating in the genome, with signatures linked to education and smoking appearing stronger among Americans born more recently. The research, published in Behavior Genetics, suggests that the genetic consequences of partner matching may be growing in the United States across twentieth-century birth cohorts.
Assortative mating refers to the non-random pairing of individuals with similar characteristics, such as schooling, body measurements, or daily habits. Over generations, this pattern can concentrate particular traits within families and alter how inherited traits are distributed across populations. The study focuses on how these dynamics leave enduring, genome-wide traces.
To explore these questions, Gretchen R. B. Saunders of the University of Minnesota analysed large genetic datasets from two major resources: the UK Biobank in Britain and the All of Us Research Program in the United States. The sample included 82,398 likely spouse or long-term cohabiting opposite-sex pairs and 16,273 full siblings from the UK Biobank, alongside 333,491 unrelated individuals from the UK Biobank and 228,875 unrelated individuals from All of Us. The researcher restricted all samples to individuals of European genetic ancestry to ensure statistical accuracy.
Eight traits were examined, spanning educational attainment, height, body mass index, and five smoking and drinking behaviours. For each participant, polygenic scores were constructed—numerical measures that sum thousands of tiny DNA variations to estimate genetic predisposition for a trait.
The study employed several approaches to track mating patterns across generations. It compared direct similarities in observable traits and in polygenic scores between spouses in the UK Biobank, and then looked for elevated polygenic-score similarity among full siblings. It also analysed correlations between even-numbered and odd-numbered chromosomes in unrelated individuals to detect broad genomic signatures of assortment and tested whether these signatures shifted across birth years.
Results showed positive similarities between spouses for all eight measures. Educational attainment exhibited the strongest spousal correlation at 0.45, followed by weekly alcohol consumption at 0.42. Moderate similarities appeared for smoking cessation at 0.33, BMI at 0.26, and smoking initiation at 0.23. The data also pointed to two distinct trait clusters: those with higher educational attainment tended to partner with taller individuals, lower BMI and lower smoking rates, while heavier engagement in smoking or drinking tended to pair with partners who shared similar substance-use patterns and higher BMI.
Genetically, the pattern persisted but at smaller magnitudes. The strongest spousal genetic correlation was for educational attainment (0.16), with height at 0.08 and smoking initiation at 0.06. Comparisons among full siblings supported the existence of assortative mating in earlier generations, as siblings showed more genetic similarity than the 0.50 expectation under random mating for most traits.
On average, genetic correlations between partners were about 80 per cent smaller than the correlations observed for the traits themselves. Part of this gap is expected because polygenic scores capture only a portion of genetic influences, but the disparity rose above 90 per cent for habits that can change within a relationship, such as drinks per week and cigarettes per day. Saunders interprets this as evidence that couples may become more alike through shared environments and mutual influence, while traits set before people meet, such as height and education, show smaller gaps.
Further analysis comparing spouse and sibling pairs suggested that genetic assortment was largely stable or even expanding from the parental generation to the offspring generation. Traits including educational attainment, height and smoking initiation displayed increasing genetic resemblance between partners relative to the excess similarity seen among siblings.
In the analyses of unrelated individuals, the even-odd chromosome method yielded persistent genome-wide signatures of assortment. Across both cohorts, educational attainment showed an even-odd chromosome correlation of about 0.07, with height and smoking initiation around 0.04. Under random mating, such correlations would be zero, indicating genuine Genome-wide patterns of partner matching.
Birth-cohort trends differed between the two countries. In the UK Biobank group (born 1936–1970), the even-odd measure remained largely stable over decades, increasing only for smoking cessation. In All of Us, which includes people born from 1900 to 2005, genetic assortment rose across generations for smoking initiation, cessation and educational attainment, with the education signal climbing from roughly 0.04 in the 1930s birth cohort to about 0.09 around 2000.
Saunders notes that the stronger trends in the United States may partly reflect All of Us’ wider birth-year range, which makes small shifts easier to detect, as well as differences in social change. She points to more rapid expansion of higher education in the United States and sharper social divides in smoking decline as possible factors.
The findings prompted cautious commentary from experts outside the study. Norbert Meskó, a professor at the University of Pécs and director of the Evolutionary Psychology Lab, said the study provides credible evidence that people resemble their partners in education, smoking, and alcohol use, with smaller but consistent genetic similarities in these traits. He emphasised that the study does not prove people today are choosing more similar partners, since historical mating patterns are inferred from genetic data rather than direct observations.
“The main takeaway is that patterns of similarity between partners can leave detectable genetic signatures across generations,” he told PsyPost. “People resemble their partners in education, smoking, and alcohol consumption, and this study also finds smaller but consistent similarities in genetic scores associated with these characteristics.”
Meskó also cautioned against over-interpreting the results. “The finding that deserves particular attention is that some of these genetic signatures were stronger in more recent U.S. birth cohorts, especially for educational attainment and smoking-related traits,” he said. “This suggests that the accumulated genetic consequences of assortative mating may differ across generations. However, I would distinguish this finding from direct evidence that people today are choosing more similar partners: the study primarily infers historical mating patterns from genetic data.”
He stressed that there is no suggestion that genes determine who one falls in love with, and noted that polygenic scores are statistical predictors, not fixed potential. “It is a statistical predictor derived from associations in particular populations and environments,” he added. “The implications for inequality should be treated as plausible consequences rather than outcomes directly demonstrated here.”
Meskó also called for longitudinal research that measures people before they form relationships and follows both partners over time, to disentangle three processes: choosing someone already similar, meeting through shared environments, and becoming more similar through life together. He concluded that such studies, combining genetic data with generations of information, would strengthen the historical interpretation.
Saunders’ study, entitled Changing Patterns of Assortative Mating for Educational, Substance Use, and Anthropometric Traits, adds a nuanced picture of how partner similarity can echo in genetic data and how those echoes may shift across generations.
