Brains or beauty? Causal evidence on the returns to education and attractiveness in the online dating market

https://doi.org/10.1016/j.jpubeco.2021.104372Get rights and content

Abstract

We study partner preferences for education and attractiveness by conducting a field experiment in a large online dating market. Fictitious profiles with manipulated levels of education and photo attractiveness send random invitations for a serious relationship to real online daters. We find that men and women prefer attractive over unattractive profiles, regardless of own attractiveness. We also find that high-educated men prefer low-educated over high-educated profiles as much as high-educated women prefer high-educated over low-educated profiles. With preferences similar for attractiveness but opposite for education, two groups are more likely to stay single: unattractive, low-educated men and unattractive, high-educated women.

Introduction

Few choices in life are as important and complex as the choice of partner. This choice involves, among other things, the partner preferences of oneself and others, which are difficult to observe. Our purpose here is to measure the underlying partner preferences that shape partner matches. In particular, we study how heterosexual men and women value brains (as measured by educational attainment levels) and beauty (as measured by facial attractiveness ratings) in a partner.

There are two good reasons to focus on partner preferences. First, social scientists (of all disciplines) have developed various theories linking partner traits to partner matches through different partner preferences. Partner matching on education and attractiveness is a case in point. These traits are on their own highly correlated between partners, much more so than most other traits (Kalmijn, 1998), and together often thought of as substitutes, with more educated men partnering less educated but more attractive women as a popular example (Becker, 1981, Buss, 1989, Jackson, 1992, McClintock, 2014).1 These matching patterns can, however, be the result of any of the three most prominent families of partner preferences—homophilic preferences, universal preferences, and traditional preferences.2 Our attempt to identify partner preferences for brains and beauty should shed light on the preferences proposed in the literature.

Second, from a welfare and policy perspective, it matters whether partner matches are driven by preferences or by constraints. If individuals, for instance, have preferences for partners with traits similar to their own (as opposed to simply being more likely to meet such partners), it may be more acceptable that positive assortative matching leads to greater inequality in family income (Eika et al., 2019). Policies aimed at reducing inequality would then incur some welfare losses.

It has proven difficult to credibly identify partner preferences. One difficulty arises because of search frictions. Individuals generally self-select into clubs, schools, jobs, and neighborhoods. If search frictions make individuals more likely to meet their partner in self-selected environments (Kalmijn, 1998), they are also more likely to end up having a partner with traits similar to their own, regardless of partner preferences. Another difficulty arises because of psychological and social frictions. Partner choices are based on a combination of rationality and emotions (Finkel et al., 2012). If fear of rejection leads disappointment averse individuals to act strategically, they may shy away from partners they consider attractive but unattainable (Bell, 1985, Loomes and Sugden, 1986, Gul, 1991). In fact, psychological research has documented that (some) people are indeed rejection sensitive, which ultimately affects their search for partners and intimate relationships (Downey and Feldman, 1996, Downey et al., 1998). Importantly, because of the fear of rejection, realized partner matches will not necessarily reflect partner preferences, even in the absence of search frictions. A final difficulty arises because of correlated traits. Any observable trait similarity between partners could be due to similarities in other related traits that are observable to potential partners but not to researchers. Whenever observable and unobservable traits are correlated, either between or within partners, the estimated partner preferences for specific traits will be biased.

In this paper, we aim to identify partner preferences by conducting the first correspondence test experiment in an authentic dating market. Specifically, we post fictitious profiles on one of the largest online dating sites in the Netherlands with traits that are manipulated along two dimensions: attractiveness and education. By randomly assigning the education level and attractiveness of our fictitious partners, we can be assured that these traits are uncorrelated with other characteristics listed in the profile, thus removing the confound of correlated traits. We focus on attractiveness (as measured by facial attractiveness) and education (as measured by educational levels) because these two traits are stable, mostly realized when online daters enter the online dating market, often considered substitutes in partner choice, a proxy for income potential (especially education), and generally highly correlated between partners. From these fictitious profiles, we then randomly sent out dating offers to a relevant set of online daters and collected their responses. By initiating the contact, we reduce search frictions as well as the fear of being rejected, which allows us to estimate unbiased partner preferences under a minimal set of assumptions, as the responses should reflect genuine preferences for partner attractiveness and education. In total, we posted 12 fictitious profiles and invited some 2,700 online daters for a date in the spring and fall of 2016.

We preview three of our main findings. First, we find that response rates significantly rise with profile attractiveness. The positive impact of attractiveness on the likelihood to respond is quantitatively large, identical for men and women, and insensitive to the attractiveness of online daters themselves. Second, we find that men are more likely to respond to lower educated profiles, but not women. The gender difference in response rates to profile education is quantitatively large, and with a pronounced asymmetry for the more educated online daters; according to our estimates, university educated men show a clear preference for lower educated women, while university educated women reject men with lower education than themselves. Third, we find no evidence that attractiveness and education are substitutable, and thus tradable, traits in response rates. The interacted impact of attractiveness and education is generally small and statistically insignificant.

As in other correspondence test studies, our main outcome is a rather crude measure that identifies the effects in the early stages of the matching procedure. We measure preferences for profile characteristics (through online responses) rather than partner characteristics. Nonetheless, the preferences we estimate (when placed into a partner matching context) predict positive assortative matching on attractiveness (so that the most attractive man matches with the most attractive woman, and so on) and negative assortative matching on education (so that the most educated man matches with the least educated woman, and so on). These predicted sorting patterns imply the persistency of traditional matches: men and women mutually prefer to live in a traditional relationship where husbands have a stronger earnings potential than their wives. These predicted sorting patterns also imply that not everyone will find a partner: the two groups at risk are unattractive, low-educated men and unattractive, high-educated women. Interestingly, these predicted sorting patterns based on responses of online daters appear to have validity outside the online dating market, which we illustrate with some supportive real-world evidence on those who get married and those who stay single.

The remainder of our study proceeds as follows. Section 2 describes earlier work on the current topic and discusses the contributions and limitations of our work. Section 3 provides details about the experiment. Section 4 outlines the main results, including results from heterogeneity analyses. Section 5 provides several robustness checks. Section 6 interprets the results in terms of partner preferences. Section 7 concludes.

Section snippets

Related literature

So far, the economics literature on partner choice has mostly estimated structural marriage models using data on realized partner matches (see, for instance, Wong, 2003, Bisin et al., 2004, Choo and Siow, 2006, Gautier et al., 2010). It is only recently that economists have begun to explore alternative strategies and data sources in attempts to extract partner preferences and their impact on partner matches, including revealed partner choice data from speed dating events and online dating sites.

Experiment

We employ a correspondence test to identify partner preferences. This is the common method to detect discriminatory preferences in the context of the labor market (Bertrand and Mullainathan, 2004), the housing market (Ahmed and Hammarstedt, 2008) and the second-hand market for customer products (Doleac and Stein, 2013). In the context of the online dating market, the method involves letting fictitious profiles with different (combinations of) traits randomly contact potential dating partners.6

Results

In this section we present causal evidence on the returns to attractiveness and education in online dating, using both figures and tables. The results are divided by gender. The figures visualize the average response rates to invitations from low-attractive, medium-attractive, and high-attractive profiles, as well as from low-educated and high-educated profiles. The tables report estimates from simple linear probability models regressing the response on profile attractiveness and education,

Robustness

While our procedure to infer partner preferences from responses to random online dating invitations has clear advantages, it goes without saying that there are disadvantages too that complicate inference. In this section, we identify several scenarios in which online responses may misrepresent preferences, and (try to) test their empirical relevance. Table 7 contains the test results; Panel A shows the results for female online daters, and Panel B for male online daters.

Are results driven by

Extrapolating results

In this section, we take one hazardous step forward, by interpreting our results on profile preferences in terms of partner preferences, and put these preferences into the context of partner matches and mismatches.

Conclusion

Few individual decisions are as important as the choice of partner. Partner choice is a constrained choice representing an unknown blend of partner preferences and constraints. Our purpose here is to measure partner preferences and particularly examine how individuals value attractiveness and education in a partner. We do this by running a large online dating experiment in which fictitious profiles with manipulated levels of attractiveness and education randomly invite real online daters for a

References (68)

  • Bell, D., 1985. Disappointment in decision making under uncertainty. Operations Res. 33(1),...
  • M. Belot et al.

    Dating preferences and meeting opportunities in mate choice decisions

    J. Human Resour.

    (2013)
  • M. Bertrand et al.

    Are Emily and Greg more employable than Lakisha and Jamal? A field experiment on labor market discrimination

    Am. Econ. Rev.

    (2004)
  • M. Bertrand et al.

    Gender identity and relative income within households

    Quart. J. Econ.

    (2015)
  • J. Birger

    Date-onomics: How Dating Became a Lopsided Numbers Game

    (2015)
  • A. Bisin et al.

    Religious intermarriage and socialization in the United States

    J. Political Econ.

    (2004)
  • M.P. Bitler et al.

    Utilization of infertility treatments: the effects of insurance mandates

    Demography

    (2012)
  • M. Browning et al.

    Economics of the Family

    (2014)
  • M. Bruhn et al.

    In pursuit of balance: randomization in practice in development field experiments

    Am. Econ. J.: Appl. Econ.

    (2009)
  • L. Bursztyn et al.

    ‘Acting wife’: marriage market incentives and labor market investments

    Am. Econ. Rev.

    (2017)
  • D.M. Buss

    Sex differences in human mate preferences: evolutionary hypotheses tested in 37 cultures

    Behav. Brain Sci.

    (1989)
  • D.M. Buss

    The Evolution of Desire

    (1994)
  • D.M. Buss

    Strategies of human mating

    Psychol. Topics

    (2006)
  • D.M. Buss et al.

    Mate preferences and their behavioral manifestations

    Annu. Rev. Psychol.

    (2019)
  • P.A. Chiappori et al.

    Fatter attraction: anthropometric and socioeconomic matching on the marriage market

    J. Political Econ.

    (2012)
  • E. Choo et al.

    Who marries whom and why

    J. Political Econ.

    (2006)
  • R.D. Clark et al.

    Gender differences in receptivity to sexual offers

    J. Psychol. Human Sexuality

    (1989)
  • J.L. Doleac et al.

    The Visible Hand: Race and Online Market Outcomes

    Economic Journal

    (2013)
  • G. Downey et al.

    Implications of rejection sensitivity for intimate relationships

    J. Personality Social Psychol.

    (1996)
  • G. Downey et al.

    The self-fulfilling prophecy in close relationships: rejection sensitivity and rejection by romantic partners

    J. Personality Social Psychol.

    (1998)
  • A.H. Eagly et al.

    The origins of sex differences in human behavior: evolved dispositions versus social roles

    Am. Psychol.

    (1999)
  • L. Eika et al.

    Educational assortative mating and household inequality

    J. Political Econ.

    (2019)
  • R. Fernandez et al.

    Love and money: a theoretical and empirical analysis of household sorting and inequality

    Quart. J. Econ.

    (2005)
  • E.J. Finkel et al.

    Speed-dating as an invaluable tool for studying romantic attraction: a methodological primer

    Personal Relationships

    (2007)
  • We thank Ingvild Almås, Dan Olof Rooth, and Adriaan Soetevent, and seminar and conference participants in Copenhagen, Glasgow, Gothenburg, Lofoten, Milan, and Nuremberg for their comments and suggestions. We also thank Tim van der Weert for his outstanding research assistance. Egebark acknowledges financial support from Jan Wallanders och Tom Hedelius stiftelse. Ekström acknowledges support from the Research Council of Norway through its Centres of Excellence Scheme, FAIR project No 262675. This research has been realized without cooperation with the online dating platform. The research project has been approved by the ethical review board at the Amsterdam School of Economics. The document with the authorized approval is made available to this journal.

    View full text