Skip to main content

A data-driven, multivariate approach to human mate preferences.

Periodic Reporting for period 1 - MULTIPREF (A data-driven, multivariate approach to human mate preferences.)

Reporting period: 2017-01-01 to 2018-12-31

Darwinian evolution is often described as ‘survival of the fittest’, but possibly an even more important evolutionary process is sexual selection. A fundamental component of sexual selection is choice of mates. In humans, studying mate preferences can provide important insight into how we engage in social and sexual relationships. However, to date research in this area typically is based on sexual selection models derived from studies of non-human species (often characterised by intense competition between males for choosy females, and the lack of formation of stable pair bonds) to identify characteristics that may be important for human mate preferences. This is problematic because this approach does not reflect human mate choice in reality, which is characterised by mutual mate choice, extended courtship, and formation of stable-pair bonds. To address these limitations, the proposed research project used a data-driven approach to identify characteristics important to human mate preferences that entirely avoids the problem of selecting candidate characteristics on inappropriate theoretical models.
This project consisted of four studies. The first three studies aimed to identify key traits that influenced preferences for faces, bodies, and personality traits, respectively, while study four aimed to integrate these results and develop a multivariate model of human mate preferences.
In Study 1, with a sample of 594 facial images, we tested the predictive utility of existing theory-driven models of facial attractiveness compared to a data-driven, statistical model. Scores for facial traits identified by previous theory were calculated on both shape and colour dimensions, including facial averageness, sexual dimorphism, asymmetry, BMI correlates, and representational sparseness. Our results showed that the data-driven model reliably explained significantly more variance in attractiveness than the theory-driven models. These results present important new evidence for the utility of data-driven approaches to studying facial attractiveness and help identify facial traits associated with facial attractiveness previously undiscovered by theory-driven research. These include traits such as face elongation, the ratio of feature size to face size, skin tone, and feature contrast. This manuscript is currently under review for publication, and a pre-print is available at

Study 2 aimed to identify body dimensions important to mate preferences using a data-driven approach. We have collected a large sample of 10,024 images of individuals from online dating websites. For each image, using a feature detection algorithm developed in the computer sciences, and Latent Dirichelt Allocation (LDA), a clustering algorithm that can be used to detect commonly occurring objects in a sample of images. We will be able to determine the common objects individuals include in their image when attempting to attract a partner. I will be able to assess patterns of objects among participants to determine which may be important under certain contexts. It is estimated this analysis should be completed within the next month. In addition, a separate sample of 99 participants have rated the images on attractiveness. From these ratings, I will be able to test patterns of universal preferences for the objects identified by the LDA.

Study 3 aimed to identify personality dimensions important to human sexual selection using a data-driven approach. The protocol for this study was designed to as closely parallel the protocol for the Study 2 as possible. First, I collected a large sample of written descriptions from online personal advertisements (N = 7,973). I then conducted an LDA on the word occurrences in each profile, which identified 25 common topics that individuals use when attempting to attract a partner online. Findings include men being more likely to advertise education/status, while women were more likely to discuss being honest/nurturing and caring for pets. We also collected ratings of desirability on a subset of these written descriptions (N = 468) from a separate group of 100 participants in the lab. Overall, both male and female profiles that discussed enjoying outdoor activities, and music and/or art were rated as more desirable. Women that discussed a healthy/active lifestyle or mentioned friends/family were also rated as more desirable. Both men and women who discussed sex or mentioned being a parent were rated as less desirable. The manuscript has been accepted for publication in Evolution and Human Behavior (preprint available at

Finally, Study 4 consisted of a large iterated study aimed at integrating the separate characteristics identified from the previous studies. I have developed the online platform to administer the study to participants, which is an application that records attractiveness ratings of stimuli and procedural generates new stimuli based on previous responses. This essentially mimicks Darwinian evolution mechanisms, where traits rated as attractive are selected for and become more frequent in the population of stimuli, while traits rated as unattractive are selected against and become less frequent. What we should expect is that traits that are priorities when assessing potential partners would be heavily selected for at the start of the experiment, while traits associated with attractiveness but are less important would face selection at later trials.
This project has gone beyond the current state of the art in two important ways. First, it is the first project to use a data-driven approach to identity the key characteristics for human mate preferences. Second, it is the first to develop a comprehensive multivariate model of human mate preferences. Previous research has been unable to investigate complex, multivariate effects due to limitations in statistical techniques and computational power. Such statistical techniques are well established in the computer science literature, but are seldom used in the field of psychology. By applying these statistical techniques, we provide unique insight into the structure of human mate preferences.

The results of this study are of enormous value for researchers in numerous fields, including evolutionary psychology, social psychology, and evolutionary biology. The vast existing literature on human mate preferences has typically examined only the effects of single characteristics identified as potentially important based on theoretical models derived from species with mating systems that are qualitatively different from those common in humans. Importantly, recent empirical studies have highlighted how this approach can lead to erroneous conclusions about human mate preferences. This project directly addresses these difficulties for the field by using new methodological approaches not subject to these limitations and generating a new model that will drive further research in the field.