Partners resemble more to one another than randomly sampled pairs. For example, partners tend to be similar with respect to various anthropometric measures (body mass index, height), socioeconomic factors, behaviours (religious views and social attitudes), lifestyle (diet, smoking habits, hobbies), and even disease risk. However, it is currently unclear what exactly causes this similarity. It is known that the major contributor to the phenotypic similarity is partner choice: people actively look for partners who are similar to them, a phenomenon known as assortative mating. However, once couples are formed partners influence each other and become increasingly similar over time. This latter phenomenon leads to partners converge further as they age.
There are however other, indirect, reasons for partners being more similar. Confounding factors are social, environmental and other components that equally impact the characteristics of partners. For example, people sharing geographic location are not only more likely to form a couple, but due to the shared environment they may be exposed to similar air pollution levels, they may have easier access to university or certain sport facilities. Such couples are expected to be more similar with respect to lung diseases, education level, or physical activity levels, respectively. Such similarity is not due to choice, but due to confounding. More broadly, one has to distinguish direct and indirect assortment. Similarity for a (secondary) trait is due to indirect assortment when mate choice occurs for a primary trait, which is correlated or causal for the secondary trait. For example, partner similarity for body mass index (BMI) is partly due to a primary mate choice for educational attainment, which impacts BMI and hence increases similarity in terms of BMI as well.
While couple correlations for a wide range of phenotypes have been explored, the contribution of the abovementioned phenomena to this apparent partner similarity is largely unexplored. In this work, we used Mendelian Randomisation (MR) to direct between-partner assortment (causality) in ~52,000 couples for 118 traits in the UK Biobank. Note that this concept is different from classical MR designs, where both exposure and outcome traits are measured in the same individual. Here, the exposure and outcome traits belong to different people.
We found evidence of partner choice for 64 (out of the 118) traits, 40 of which had larger phenotypic couple correlation than causal effect. This suggests that confounders contribute to trait similarity. When we systematically searched for key confounders, identified household income, overall health rating, and education levels. These confounders accounted for 29.8, 14.1 and 11.6% of the observed correlations between partners, respectively. It implies that, on average, almost a third of the observed correlation between partner traits can be explained by their shared socio-economic status. We also quantified the contribution of indirect assortment for each of 118 phenotypes. For example we observed that the couple correlation () for systolic blood pressure (SBP) is in part driven by direct assortment () and partly due to primary assortment for obesity and physical activity, each of which having a causal impact on SBP, giving rise to inflated couple correlation driven by indirect assortment.
We observed no differences between male-to-female vs female-to-male assortment. On the other hand, partner similarity for BMI decreases as couples live longer together, but their smoking status is converging over time.
Finally, our mediation analysis revealed that most causal associations between different traits in the two partners are indirect. For example, the apparent causal effect of TV watching of an individual (Xindex) on the BMI of its partner (Ypartner) is due to primary assortment for TV watching between partners combined with the impact of increased TV watching leading to increased BMI within the same individual.
In summary, our results illustrate the different mechanisms contributing to the similarity between partners. These analyses revealed how direct assortment combined with confounding and indirect assortment can give rise to the observed partner similarity.