Do birds of a feather flock together, or do opposites attract? These are both common aphorisms, which means that they are commonly offered as generally accepted truths, but also that they may by wrong. People like pithy phrases, so they spread prolifically, but that does not mean they contain any truth. Further, our natural instincts are not adequate to properly address whether they are true or not.

Often people will resort to the “availability heuristic” when confronted with these types of claims. If they can readily think of an example that seems to support the claim, then they accept it as probably true. We use the availability of an example as a proxy for data, but it’s a very bad proxy. What we really need to address such questions is often statistics, something which is not very intuitive for most people.

Of course, that’s where science comes in. Science is a formal system we use to supplement our intuition, to come to more reliable conclusions about the nature of reality. Recently researchers published a very large review of data, a meta-analysis, combined with a new data analysis to address this very question. First, we need operationalize the question, to put it in a form that is precise and amenable to objective data. If we look at couples, how similar or different are they? To get even more precise, we need to identify specific traits that can be measured or quantified in some way and compare them.

For the meta-analysis part of the study the authors looked at “480 partner correlations from 199 peer-reviewed studies of co-parents, engaged pairs, married pairs and/or cohabitating pairs” between 1903 and 2022, including millions of couples. They limited this analysis to opposite-sex couples and plan to extend their analysis to same-sex couples in a later analysis. They then did their own analysis of 133 traits in 79,074 couples using the UK Biobank database. Correlations are expressed from -1 to 1, with -1 meaning that traits were always opposite, 0 meaning no correlation at all, and 1 meaning that traits were always the same within a couple. For example, if introverts always coupled with extroverts that would equal a negative correlation of -1. If extroverts always coupled with extroverts, that would be a correlation of 1. If the probability of an extrovert and introvert getting together were the same as the background population, that would be the absence of any statistical correlation, and therefore 0.

What the researchers found is generally a positive correlation among the studied traits – so taken as a whole, opposites do not attract, but rather people tend to gravitate toward people who are more similar to themselves than the general population. There was a range of correlations. The strongest correlation (0.87) was for year of birth, which makes sense as people tend to couple with people of similar age, and school tends to put people of the same age together in social groups.

There were also high correlations for political views, religious views, education level, and substance use. This fits with prior research (unsurprisingly) which has found some degree of “assortative mating”. What this means is that society tends to sort people into social circles based on background and socio-economic status, and that people tend to find mates within their social circle. Some researchers warn this can exacerbate over generations income and opportunity inequality. Parents tend to transmit advantages to their children, and these advantages tend to compound over generations through assortative mating.

Some physical traits, such as height, also have a positive correlation. This has an impact on studies of population genetics, which may falsely assume random mating. What this can mean practically is that more extreme traits can emerge over time – if the tallest people tend to marry the tallest people, they can have really tall children.

Of the 133 traits studies, only three had a negative correlation and they were weak. The strongest negative correlation was chronotype (being a morning person or a night owl), at -0.18. There was an even smaller negative correlations for neuroticism and hearing difficulty. I suspect these small negative correlations are likely statistical flukes and would guess that they would not replicate with independent data. But even if these effects are real for some reason, they are tiny.

The next level question is – what are the causes of the various positive correlations? I already mentioned assortative mating, but how much does this explain? Specifically, what are the primary factors and what are the downstream consequences? In other words, are there specific traits that some people are choosing to some degree,  and are there some traits that are just statistically coming along for the ride? For example, I would imagine that many non-smokers would not couple with a smoker (it’s easy for me to imagine because I am one of those people). That is an actual factor of “attraction”.

At the same time, smoking correlates with many other factors as well. For example, it is negatively associated with educational level and socioeconomic status. So even if there weren’t an attraction-factor to smoking, there would still likely be a positive correlation to this trait resulting from a separate correlation with education level and socioeconomic status. With 133 studied traits, this can become an impossible tangle of causes and effects. To make matters more complex, the arrow of causation may be reversed for some traits – meaning that as couples spend time together they may become more similar in certain traits. Perhaps their political views will tend to align over time through shared life experience and extensive conversations.

Despite these complexities, which all deserve further research and exploration, we can make some general statements. It seems that the notion that “opposites attract” is basically bunk. We should just ditch this aphorism. In reality, opposite sex couples tend to be more similar to each other in a host of ways than the general population. This has implications for several areas of research, and may have societal implications as well.