Evolutionary Psychology's Limits: Why Dating Apps Shouldn't Rely on It

The DII Take

Evolutionary psychology provides a useful but overextended framework for understanding dating behaviour. The broad predictions—that physical attractiveness matters in initial attraction, that earning potential influences long-term partner evaluation—receive empirical support. The specific, gender-differentiated predictions—that men care primarily about appearance and women care primarily about status—are far less supported by modern data than the popular version of the theory claims. Research by Eastwick and Finkel using speed-dating paradigms found that gender differences in stated mate preferences (what people say they want) do not predict gender differences in actual romantic interest (who people are actually attracted to). Dating platforms would be poorly served by designing exclusively around evolutionary psychology's predictions, which reflect stated preferences rather than revealed ones.

Where the Evidence Supports the Framework

Several evolutionary predictions receive consistent empirical support. Physical attractiveness matters for initial attraction, and this is not gender-specific. Both men and women respond strongly to physical attractiveness in initial encounters, as demonstrated by meta-analyses and speed-dating research. The evolutionary claim that men weight attractiveness more heavily than women is supported in surveys of stated preferences but not consistently in behavioural studies of actual partner choice.

Age preferences follow predicted patterns to a degree. Men do tend to prefer younger partners, and women tend to prefer slightly older partners, across most cultures studied. However, the magnitude of these preferences varies considerably by cultural context, and actual partner age gaps have been narrowing in recent decades. U.S. Census data shows the average spousal age gap declining from 4.9 years in 1880 to 2.2 years in 2022.

Where the Evidence Challenges the Framework

Several evolutionary predictions fare poorly against modern data. The earning potential gender difference is inconsistent in behavioural data. Eastwick and Finkel's speed-dating research found that whilst women reported valuing earning potential more than men in pre-interaction surveys, both genders showed similar patterns of actual attraction during live encounters. The stated-revealed preference gap is substantial and undermines the predictive utility of evolutionary models in dating product design.

The most commercially important finding from the evolutionary psychology literature is not about evolution at all. It is the gap between what people say they want in a partner and who they are actually attracted to.

Cultural variation exceeds what a universal evolutionary framework predicts. Whilst Buss's 1989 study found consistent cross-cultural patterns in some preferences, subsequent research has documented enormous variation in the strength and direction of mate preferences across cultures. The 2024 Eastwick et al. registered report, testing ideal partner preference matching across 43 countries, found significant but small effects that varied substantially by context.

Same-sex mate preferences present challenges for a framework built on reproductive fitness maximisation. The growing visibility and inclusion of LGBTQ+ users on dating platforms requires preference models that cannot be derived from heterosexual reproductive logic. For dating operators, the practical implication is that platform design should not assume fixed, gender-binary preference patterns. Flexible filtering systems, personalised recommendation models that learn from revealed rather than stated preferences, and inclusive design that accommodates diverse preference patterns will serve users better than systems built on evolutionary psychology stereotypes.

Product Design Implications

The evolutionary psychology debate has practical implications for platform design. Flexible preference systems outperform rigid evolutionary assumptions. A platform that learns individual preferences from revealed behaviour produces recommendations reflecting the complexity of human attraction. The Eastwick and Finkel research strongly supports this: stated preferences are unreliable guides to actual romantic interest.

Age filter defaults deserve scrutiny. Most platforms allow age filtering, and defaults often reflect evolutionary assumptions. Nudging users to expand age ranges, as Hinge has begun doing, may increase match rates by connecting users with compatible partners they would have filtered out. The diversity of human attraction patterns far exceeds what any single theoretical framework can capture. Platforms designing for this diversity will serve the broadest possible user base. The most commercially successful dating products will be those that accommodate individual variation rather than imposing population-level stereotypes on individual matching.

The Stated-Revealed Preference Gap

The most commercially important finding from the evolutionary psychology literature is not about evolution at all. It is the gap between what people say they want in a partner and who they are actually attracted to. Research by Eastwick and Finkel (2008), using speed-dating paradigms, demonstrated this gap experimentally. Before speed-dating events, participants rated the importance of various partner attributes. Men rated physical attractiveness as more important than women did, and women rated earning potential as more important than men did—consistent with evolutionary predictions. But when the researchers examined who participants were actually attracted to during the speed-dating events, these gender differences disappeared. Both men and women were attracted to physically attractive partners, and both showed similar (weak) responses to earning potential cues.

Algorithms that rely on stated preferences to generate matches are building on a foundation the research shows is unreliable. An algorithm that learns from revealed preferences will produce more accurate recommendations.

This finding has been replicated across multiple studies and has profound implications for dating algorithm design. Algorithms that rely on stated preferences to generate matches are building on a foundation the research shows is unreliable. An algorithm that learns from revealed preferences—observing who a user actually engages with, messages, and meets, rather than what they say they want—will produce more accurate recommendations.

The practical challenge is that revealed preference data takes time to accumulate. New users have no behavioural history for the algorithm to learn from, forcing a reliance on stated preferences during the critical early engagement period. This 'cold start problem' is universal in recommendation systems but particularly consequential in dating, where early dissatisfaction drives rapid churn. A hybrid approach—using stated preferences as an initial filter whilst progressively incorporating revealed preference data—represents the most evidence-supported algorithm design.

Beyond Binary Gender

The evolutionary psychology framework was built on a heterosexual, binary-gender model that increasingly fails to describe the diversity of modern dating populations. LGBTQ+ mate preferences do not follow the gender-differentiated patterns that the framework predicts, and the growing representation of non-binary, genderqueer, and gender-fluid individuals on dating platforms requires preference models that cannot be derived from binary reproductive logic.

This limitation is not merely theoretical. Platforms that design their recommendation algorithms around evolutionary psychology's gender-binary predictions may systematically underserve LGBTQ+ users by applying preference assumptions that do not reflect their actual attraction patterns. The most inclusive approach is to learn each user's preferences individually from their behaviour, without imposing any theoretical framework on what those preferences 'should' look like.

The Speed-Dating Revolution in Mate Preference Research

Speed-dating methodology, pioneered in academic research by Eli Finkel and Paul Eastwick, transformed the study of mate preferences by allowing researchers to compare what people say they want with who they actually pursue. The results were transformative for the field. In a landmark 2008 study published in the Journal of Personality and Social Psychology, Eastwick and Finkel had speed-dating participants rate the importance of various partner attributes before the event. Women rated earning potential as more important than men did, and men rated physical attractiveness as more important than women did—consistent with evolutionary predictions. But when the researchers examined who participants actually selected for follow-up dates, these gender differences vanished. Both men and women were primarily influenced by physical attractiveness, and neither showed the sex-differentiated preference for earning potential that their pre-event ratings predicted.

This stated-revealed preference gap has been replicated across dozens of speed-dating studies worldwide. It represents one of the most robust findings in modern mate preference research and has profound implications for dating platform design. Algorithms that rely on stated preferences (what users say they want in their filter settings) are building on a foundation that the research shows is unreliable. An algorithm that learns from revealed preferences (observing who users actually engage with) will produce more accurate recommendations.

The 2024 Eastwick, Sparks, Finkel et al. registered report, testing ideal partner preference matching across 43 countries with over 10,000 participants, found that preference-matching effects were statistically significant but small. People do, to a modest degree, evaluate partners more favourably when those partners match their stated ideals. But the effect sizes are far smaller than popular belief and dating platform marketing would suggest. The practical implication is that matching on stated preferences adds some value but should not be the primary basis for algorithmic recommendation.

The Cultural Relativity Challenge

David Buss's original 1989 study of mate preferences across 37 cultures found broadly consistent gender differences, but subsequent research has revealed far more cultural variation than the original study captured. The magnitude of gender differences in mate preferences varies enormously across cultures, with the largest differences found in traditional, patriarchal societies and the smallest in egalitarian, economically developed ones. This cultural variation poses a direct challenge to the universality claims of evolutionary psychology. If gender differences in mate preferences are smaller in egalitarian societies, the differences may be driven more by social structure than by evolved psychology. The practical implication for dating platforms is that preference models calibrated on one cultural population may not generalise to another. A matching algorithm trained on American user data may make inaccurate preference assumptions when applied to Scandinavian or Japanese users.

The mate preference literature is evolving rapidly, with new large-scale, cross-cultural studies producing more nuanced findings than earlier work. For dating operators, the key takeaway is flexibility: platforms should learn individual preference patterns from behaviour rather than imposing theoretical assumptions about what men or women should want. The most successful platforms will be those that accommodate the full diversity of human attraction, which the research shows is broader, more variable, and more individually specific than any single theoretical framework predicts.

The platforms that design for human diversity rather than theoretical simplicity will serve their users best and build the most defensible products in an increasingly scrutinised market.

The debate between evolutionary and social-structural explanations for mate preferences is not merely academic. It determines how dating platforms should be designed. An evolutionary framework predicts fixed, universal, gender-differentiated preferences that algorithms can hard-code. A social-structural framework predicts flexible, culturally variable, individually diverse preferences that algorithms should learn dynamically. The accumulated evidence supports the social-structural view more than the evolutionary one, at least for the specific predictions about gender differences that dating platforms most commonly rely upon. The stated-revealed preference gap, the cultural variation in effect sizes, and the growing data on non-heterosexual mate preferences all argue for flexible, behaviour-learning algorithms over rigid, assumption-based ones. The platforms that design for human diversity rather than theoretical simplicity will serve their users best and build the most defensible products in an increasingly scrutinised market.

The evolutionary framework retains value as one input among many for understanding human mating behaviour. The broad prediction that physical attractiveness matters in initial attraction is well-supported and relevant to product design. The specific, gender-binary predictions about differential emphasis on appearance versus resources are far less supported and should not be the basis for platform design decisions. The most honest scientific assessment is that human mate preferences are shaped by a complex interaction of evolved tendencies, cultural norms, individual experience, and situational context, and that no single framework captures this complexity adequately.

More articles by DII Editorial