DECISION MAKING IN MATE CHOICE MARKETS
Stephen Whyte
BBus(Econ) with Distinction, MRes(Econ)
Principal Supervisor: Prof. Benno Torgler
Associate Supervisor: Prof. Uwe Dulleck
Associate Supervisor: Dr David A. Savage
Submitted in fulfilment of the requirements for
Doctor of Philosophy (Economics)
School of Economics & Finance
QUT Business School
Queensland University of Technology
2018
1
Abstract
This thesis contributes to the behavioural economic and behavioural science literature by providing empirical evidence of factors that impact large scale decision making in mate choice settings. The thesis consists of five studies. The first three explore human decision making in the online dating market. They utilise both stated preference and actual decision outcomes to explore differences between preference and choice, positive assortment, and specificity of preference for different sexes. Studies four and five concern male and female behaviour in the clinical and informal (online) sperm donation markets. These studies explore factors that impact female’s decision to choose specific donors, and the characteristics, personality and behaviour of males who choose to donate.
2
3
Keywords
Decision making; Mate choice; Sex differences; Online dating; Online sperm donor; Nonbinary gender; Parental investment theory; Positive assortment; Homogamy; Preference; Choice.
4
5
Acknowledgments
Thank you to Benno Torgler and Uwe Dulleck for your incredible support and encouragement across my research journey. Your generosity, knowledge, wisdom, empathy, leadership, humour and understanding has inspired me every day. No words can express how much I appreciate all that you have done for myself and my family, and how dearly I value our friendship.
To all the QUT Econ team, past and present, thank you for listening, helping, crying with, laughing at and with, drinking with, bitching and whinging, and most of all supporting me in any and every way that you did. I hope you enjoyed it, because I loved every second and I had an absolute blast working with you all. Thank you to Ben, Naomi, David, Marco, Sam, Juliana, Suzanne, AKK, Dr Mueller, Jieyang, Harry, Azhar, Jonas, Louisa, Tommy, Tony, Lionel, Megan, Jason, Gaurav, Chris, Poli, Lisa, Alice, Zili, Vlad, and anyone else I’ve missed.
To Pascalis, Michelle, Kristin, Katalina, my best buds in Finance Mary and Alice, Gareth, Lee, Diane, Brian, Takae, Maria, and any and all other QUT admin I’ve missed, thank you for all your support and help.
To all of the amazingly talented scientists who have been gracious enough to give their time in supporting me and my research along the way. Thank you to Douglas T. Kenrick, Steven Neuberg, Dan Conroy-Beam, Todd K. Shackelford, Jaimie Krems, Jordan Moon, Bill von Hippel, Karin Hammarberg, Khandis Blake, Dax Kellie, Mike Kasumovic, Amany Gouda-Vossos, Ahmed Skali, Gigi Foster, Barnaby J. Dixon, David Stadelman, and Robert C. Brooks.
Thank you to Marilyn Healy and all the Kelvin Grove QUT Human Research Ethics Team. Thank you for being the Batman to my Joker, or the Joker to my Batman, I often got confused who was playing which role in the soap opera that was my research journey. Your feedback on my work taught me so much about myself, thank you.
6
I would like to recognize and thank Lifeline for its valued role in Australian human behavioural research support, as well as acknowledge funding support from an Australian Government Research Training Program Scholarship. Thank you to all the external stakeholders who took a chance on me and helped me do something great. Erika Tranfield at PrideAngel, James Templeman at GIGA, Dave Heyson at RSVP and Keith Harrison at QFG. Without your faith in me and your passion for science I wouldn’t have had any data. Thank you for the trust and belief in my ability that you showed. Thank you to the more than 50,000 people globally that willingly provided data for the studies in this thesis. Thank you to Ash, Ethan, Will and Poppy for all your support and encouragement. I’m sorry that my PhD took so much time away from you all. I hope all my hard work has inspired you to follow your dream, no matter what that may be. I will always be your biggest supporter. I hope you believe in yourself as much as I believe in you, the future will always be bright for you.
Thank you to my mum Carmel for sticking by me through thick and thin, you never stopped believing in me.
Thank you to Diz, Avi, Leela, for all your love and support. Thank you to Ciara and Emily for being a special part of our family at a time when we needed it most.
And finally thank you to Priscilla, who is the reason I started this journey a decade ago. Thank you for helping me to realise my dream, I will always be thankful for all of your support across the last nine years, I couldn’t have done it without you.
7
Table of Contents
List of Figures 10 List of Tables 12 List of Publications 15 Statement of Authorship 16 Chapter 1: General Introduction 17 Outline of Thesis 20 Chapter 2: Things change with age: Educational assortment in online dating 22 Introduction 23 Background 24 Methods 27 Participant Pool 27 Data Collection 29 Experimental Model 29 Results 31 Discussion 38 Conclusion 40 Chapter 3: Preference vs. Choice in online dating 42 Introduction 43 Method 45 Participants 45 Procedures 46 Results 47 Discussion 52 Conclusion 54 Chapter 4: Do women know what they want? Sex differences in online daters 57 educational preferences Introduction 58 Method 59 Participants 59 Data collection and ethical practice 60 Measures 61 Preference specificity 61 Preference for homogamy or hypergamy 62 Minimum acceptable level of education 63 Empirical analysis 63 Results 64 Discussion 71
8
Chapter 5: What women want in their sperm donor: A study of more than 73 1000 women’s sperm donor selections Introduction 74 Method 77 Data 77 Results 82 Discussion 88 Limitations 90 Conclusion 93 Chapter 6: Online sperm donors: The impact of family, friends, personality 95 and risk perception on behaviour Introduction 96 Method 100 Data collection 100 Multivariate analysis 108 Results 109 Discussion 113 Chapter 7: Summary and Conclusion 119 Summary of findings 119 Limitations 121 Chapter 2, 3 & 4: Online dating 121 Chapter 5 & 6: Sperm donation in clinical and informal markets 121 Directions for future research 122 Bibliography 125 Appendices: Statement of author contribution 134
9
List of Figures
Figure 1. Age distribution 28
Figure 2. Educational hypergamy to hypogamy 30
Figure 3. Predicted margins for assortative mating, with a focus on age slopes as a 35 function of education (linear prediction). Figure 4 Predicted margins for educational hypergamy, with a focus on age slopes by 37 gender (linear prediction). Figure 5. Predicted margins for educational hypogamy, with a focus on age slopes by 37 gender (linear prediction). Figure 6. Percentage of participants contacted matching up to 7 characterist ics 47
Figure 7. Average marginal effect: Age and Gender 51
Figure 8. Average Marginal effect: Age and Education 52
Figure 9. Age distribution differentiated by sex 60
Figure 10. Percentage of sample stating an explicit educational preference 65
Figure 11. Marginal effects of sex and member’s age on the probability of stating an 68 explicit educational preference with effects of control variables reported in Table 1 taken out Figure 12. Specificity of educational preference: Pickiness 68
Figure 13. Preference for educational homogamy or hypergamy in ideal partner by 69
Figure 14. Lowest minimum acceptable educational preference in ideal mate 70
Figure 15. Frequency of sperm sample reservation (90.82% of sample) 80
Figure 16. Frequency of sperm sample reservation (49.48% of sample) 80
Figure 17. Average days elapsed between reservations (N = 1546) 81
Figure 18. Average days elapsed between reservations (49.48% of sample) 82
Figure 19. Average marginal effects: age and reservation order 87
Figure 20. Distribution of donor age 102
10
Figure 21. Distribution of donor risk perception towards informal donation 102
Figure 22a. Predicted margins for total number of women donated to 112
Figure 22b. Predicted margins for number of years informally donating 112
11
List of Tables
Table 1. Descriptive statistics on participant and contact’s educational background 29
Table 2. Positive educational assortment (based on OLS & probit regressions) 33
Table 3. Educational hypogamy and hypergamy 34
Table 4. OLS analyzing matches of stated preference and actual choice 49
Table 5. Negative binomial model analyzing matches of stated preference and actual choice 50
Table 6. Regression Coefficients from Multivariate Models of Participant Educational 66 Preference Table 7. Factors influencing number of days until reservation 84
Table 8. Interaction effects: Donor Age and reservation order 86
Table 9. Average Marginal Effects: age and reservation order 88
Table 10. Donor descriptive statistics 101
Table 11. Factors impacting number of average monthly informal donations (Nbreg) 103
Table 12. Factors impacting number of years donating informally (Nbreg) 104
Table 13. Factors impacting total number of women informally donated to in lifetime (Nbreg) 105
Table 14. Factors impacting number of offspring from donation (Nbreg) 106
Table 15. Factors impacting risk attitudes to informal donation (OLS) 107
12
List of Publications
This thesis is presented in the form of peer reviewed published papers. The thesis is comprised of the following five manuscripts.
1. Whyte, S., & Torgler, B. (2017). Things change with age: Educational assortment in online dating. Personality and Individual Differences, 109, 5-11. 2. Whyte, S., & Torgler, B. (2017). Preference versus choice in online dating. Cyberpsychology, Behavior, and Social Networking, 20(3), 150-156. 3. Whyte, S., Chan, H.F., & Torgler, B. (2018) Do women know what they want? Sex differences in online daters educational preferences. Psychological Science. forthcoming 4. Whyte, S., Torgler, B., & Harrison, K. L. (2016). What women want in their sperm donor: A study of more than 1000 women’s sperm donor selections. Economics & Human Biology, 23, 1-9. 5. Whyte, S., Savage, D. A., & Torgler, B. (2017). Online sperm donors: the impact of family, friends, personality and risk perception on behaviour. Reproductive BioMedicine Online, 35(6), 723-732.
13
Statement of Original Authorship
The work contained in this thesis has not been previously submitted to meet requirements for an award at this or any other higher education institution. To the best of my knowledge and belief, the thesis contains no material previously published (except by myself) or written by another person or where due reference is made.
Signature: QUT Verified Signature
Date: 25th January 2018
14
Chapter 1: GENERAL INTRODUCTION
Choosing a mate is arguably the largest economic decision a human can make. Mate choice decisions can have significant short and long run impacts on the individual decision maker, as well as macroeconomic impacts for society. By observing and studying individuals mating preferences and behaviours, and the factors at play when individuals make mate choice decisions, behavioural science can build a more developed understanding of the unseen mechanisms that drive large scale decision making processes. Exploring the interplay of such factors as an individual’s biology, personality, education, income, sex, gender, sexuality, religion, political beliefs and micro level behaviour, can inform and develop sciences understanding of how humans make decisions. The study of mate choice decision making then provides considerable advantages for the applied microeconomic analysis of human behaviour. The application of a more nuanced and holistic understanding to the factors at play in mate choice decision processes is the necessary foundation for future policy initiatives that assist frontline psychological healthcare and economic support services, as well as legislative frameworks that seek to combat sex, gender and sexuality based discrimination and inequity.
In many mate choice settings participants (to reduce their search costs) advertise a preference for the characteristics of the individual they are searching for. To further increase the likelihood of finding a partner, individuals also willingly provide significant amounts of their socio-demographic characteristics and discerning features. Their preferences and behaviours, and most importantly their “decisions” are often readily observable. Because one of the lowest cost and most readily accessible modern conduits in searching for a mate is the internet, a myriad of data is now readily accessible for behavioural research. These data allow
15 for the opportunity to test a range of economic, psychological, and wider social science theories using large populations of real world participant’s actual behaviour. For this reason, this thesis primarily focuses on the cyber mate choice domains of online dating and online reproductive tissue (sperm) donation.
Human males and females face different reproductive costs and constraints. For males, the cost of reproduction can be as little as the time spent in copulation. For females however internal gestation, lactation, paternity certainty and ongoing maternal investment all create a significantly disproportionate opportunity cost of reproductive market participation. Just as economics is the study of scarcity and finite resources, evolutionary psychology too recognises the costs and constraints involved in reproduction, and the evolutionary pressure it places on the decision makers involved. Triver’s (1972) seminal work titled Paternal
Investment and Sexual Selection proposed that the sex who bears the heavier reproductive burden (human females) should naturally be more selective in their mate choice behaviour.
Further behavioural research work has built on this core theory of female choice (Buss 1989;
1991) by proposing that females should not only be more selective than males, but that females should exhibit an evolved adaptive preference for male with resources, to compensate for this disproportionate opportunity cost in reproduction. Resources including not just accrued wealth, but other proxies for possible increased or ongoing paternal investment, such as education (Buss 1989; Buss & Barnes, 1986; Shackelford et al., 2005) and age (Buunk et al., 2002; Kenrick & Keefe, 1992). This thesis applies many of these core economics and evolutionary psychological normative constructs to explore key sex differences in human mate choice behaviour, and the contributing individual factors that impact the decision mechanism. By looking through an applied microeconomics lens this thesis is able to quantitatively explore and test these overlapping theories of large scale mate choice decision making, using uniquely large samples of actual human behaviour.
16
This thesis consists of five different studies presented in Chapters 2 through Chapter 7.
In Chapter 2 I provide empirical evidence of positive educational assortment in online daters contact behaviour. While previous research has shown that humans exhibit a preference for the same level of education in their partner as their own (Buss, 1985; 1989), this study goes a step further and explores educationally hypergamous and homogamous mate choice behaviour. From a neoclassical economics perspective rational maximisation (ceterus paribus) should motivate individuals to prefer/favour those with equivalent or higher levels of education relative to their own. Such online dating market data provides a uniquely observable setting in which to explore factors that impact homogenous preference (positive assortment), as well as strategic and utility maximising behaviours associated with heterogenous educational partner selection.
In Chapter 3 I provide empirical evidence that individuals ex ante preference for the characteristics of their ideal mate show little to no explanatory power in their actual contact behaviour. The online dating market provides the perfect observational setting for exploring the difference between what humans say they want, and how they behave in practise. This study utilises the ideal partner preferences individuals’ state before they begin their online search for a mate, and compares those selections with the characteristics of the individuals that they actually contact on the dating website in question. As many of our core economic, societal and legal processes heavily rely on eliciting individual’s true preference (census data, voting preferences, welfare transfers, education, healthcare, infrastructure investment, but to name a few) establishing factors that impact individual’s decision to deviate from their prior is critically important. Such research can also provide insight into the cognitive biases and constraints humans have and employ when making search or consumption decisions.
Chapter 4 presents (to the best of the authors knowledge) the first ever large scale empirical evidence of sex differentiated selectiveness in mate choice preference for different
17 age groups. While a significant amount of literature (Buss, 1989; 1991) has demonstrated females evolved adaptive preference for proxies for resources (namely, education) no study has used such a large sample to conclusively demonstrate female preferences (when compared to males) across all ages are more stringent and become increasingly more selective through the years of peak fertility. Exploring how males and females differ in the way they search and select mates can provide insight into how different sexes form and select their preference, and why individual’s preferences change across time.
Even after more than four decades of commercial sperm banking, no study has ever quantitatively analysed the characteristics of the sperm donors that females have actually selected to use in clinical invitro fertilisation (IVF) and donor insemination (DI) settings.
Chapter 5 is thus the first ever behavioural study of its kind. For the first time in human history males are (from a legal standpoint) consciously instructed not to invest in their resulting offspring in any way. Studying the impact (of zero obligation for paternal investment from the offsprings biological father) on female preference and behaviour provides a unique and unparalleled large-scale decision setting for behavioural research.
Males who donate their sperm in clinical donation settings are not seeking (or permitted) to invest in their offspring in the traditional sense (eg. contact, financially, emotionally). For millennia females have evolved to seek cues for males that will paternally invest in their offspring before they choose to reproduce with them (Trivers, 1972). As the opportunity cost of reproduction is far greater for women compared to men (eg. 9 months internal gestation,
12-18 months of lactation), identifying and securing males who could historically compensate and support this disproportionate female reproductive cost seems an obvious stipulation or criteria for their choice in a reproductive mate. Clinical sperm donation provides arguably the greatest possible natural field experiment setting to explore and delineate female mate choice psychology from male’s signals for accrued resources or those that may be proxies for
18 paternal investment. The Chapter 5 study uses a unique sample of more than a decade of women’s sperm donation reservations (selections) at an Australian fertility organisation. It uses the speed at which donor’s sperm samples were reserved (purchased) for use by women to explore female’s preference for specific characteristics in the donor males they chose for procreation.
Markets often fail to allocate efficiency. A lack of supply in the global clinical sperm donation market is one such real world example. Since IVF and commercial sperm banking began more than four decades ago, male supply has consistently fallen short of female demand. Regulatory factors that have excluded donors based on genetic testing, as well as socio-cultural and sexuality based forms of discrimination have further compounded this shortage. The commercialisation & regulated global trade in clinically donated semen
(although with good intention) has by its very nature forced men and women unable to access clinical donation opportunities to seek and create alternative options. The development of a cyber marketplace for sperm and egg donors has provided men and women with a low
(opportunity) cost alternative to the historically oligopolistic global supply of clinical reproductive tissue donation. The Chapter 6 study explores determinants of men that have made the large scale decision to participate as a sperm donor using the online connection website PrideAngel. The study provides unique empirical evidence on how males perceive the risks involved in online participation, and factors that drive success (ie. number of donations, number of women donated to, and number of offspring generated). It also explores the role and links between kinship, social networks and personality in donor psychology and behaviour. As informal (online) donation is an unregulated market facilitated from the advent of the internet, exploring individual’s decision processes and behaviour in an unregulated setting can provide insight into mechanisms and forces that trigger market development. And
19 more importantly, factors that drive individuals to participate, collaborate and act altruistically in large scale decision settings.
Outline of thesis
I present the five essays comprising this PhD thesis in individual chapters. Each chapter is self-contained and thus provides the relevant literature independently for each study. Due to the requirements set down by the Queensland University of Technology (QUT) in their “thesis by publication” requirements some minor adjustments and changes have been made from the published version of the manuscripts included. These changes primarily relate to the reformatting of tables, figures, and references and their aggregation into single lists for this thesis document.
Firstly, I present three explorative studies of online dating behaviour utilizing data from the Australian online dating website RSVP. Chapter 2 titled Things change with age:
Educational assortment in online dating is a joint work with Benno Torgler and published in
Personality and Individual Differences. Chapter 3 is another collaboration with Benno
Torgler published in Cyberpsychology, Behavior and Social Networking, its title Preference vs. Choice in Online Dating. Chapter 4 is the final study using data from RSVP and is a collaborative work with Ho Fai Chan and Benno Torgler titled Do women know what they want? Sex differences in online daters’ educational preferences.
The second section of the thesis contributes to the field of reproductive medicine and wider behavioural science by exploring sperm donor and recipient psychology and behaviour in both clinical and informal settings. Chapter 5 was a collaborative work with Benno Torgler and Keith Harrison using data from the records of Queensland Fertility Group (QFG). This
20 work is published in Economics and Human Biology and is titled What women want in their sperm donor: a study of more than 1000 women’s sperm donor selections. Chapter 6 explored donor characteristics and behaviour in the informal (unregulated) sperm donation market using data collected from the online connection website PrideAngel. This study was a joint work with David A. Savage and Benno Torgler. The study is published in Reproductive
Biomedicine Online, and is titled Online sperm donors: the impact of family, friends, personality and risk perception on behaviour.
Finally, in the Conclusion chapter I provide a summary of key findings and limitations for each study. I then conclude the thesis by providing future directions for work in these unique and innovative fields for behavioural science research.
21
Chapter 2: THINGS CHANGE WITH AGE: EDUCATIONAL ASSORTMENT IN ONLINE DATING
Stephen Whyte & Benno Torgler
Personality and Individual Differences (2017) 109, 5-11
Abstract
To identify the factors that influence educational assortment in an online dating setting, we analyse 219,013 participant contacts by 41,936 members of the Australian online dating website RSVP. Consistent with prior research, we find more educated online daters are consistently likely to assort positively (homogamy), meaning that they are more likely to contact potential mates with the same level of education. However, as the more educated cohort gets older they care less about homogamy while less educated daters become more interested in homogamy which leads to an increase in similarity towards caring for the same educational level. Older and more educated online daters are less likely to contact those with lower educational levels compared to their own, while women are more likely to contact a potential mate with higher education relative to their own (hypergamy). Our interaction analysis reveals fewer differences in educational hypergamy among older participants but a greater likelihood of online daters contacting mates with lower levels of education among younger males and older females. Further research is warranted on technology’s impact on human mating behaviour; particularly the psychology employed by humans using the Internet to maximize their chances of matching their educational preferences in a mate.
22
Mate choice is not a random roll of the dice, nor is it the search for universal signs of beauty upon which everyone agrees. There is, as the maxim goes, “a lid for every pot”. Aside from searching for those signs which everyone finds attractive, each individual is also searching for his or her own version of that perfection. That person is the one who is most like themselves.
Keller et al. 1996. p. 221
Introduction
Assortative mating (or homogamy) refers to the non-random matching of individuals based on a preference for a similar or shared feature (Buss, 1985). For over half a century, assortative mating research has identified and explored a range of characteristics and traits that individuals not only prefer but actually choose in a partner, including symmetry in factors such as age, aesthetics, attractiveness, personality, culture, education, religion and race
(Berscheid et al., 1971; Buss, 1985; Little et al., 2003; Mare, 1991; Vandenburg, 1972). More recent research has even used genome-wide genotype data to measure the magnitude to which married couples assort genetically (Guo et al., 2014, p. 2). This plethora of research underscores the importance of understanding the psychology of actual mate choice behaviour rather than mere stated preference in human mating scenarios.
The Internet has created a completely new conduit through which humans can search for a mate. It constitutes a non-sequential decision-making setting for mate choice, one that permits multiple partner choices in real time facilitating significantly greater available choice of potential mates, particularly on factors such as education that may have historically constrained the number of potential mates. Thus, not only does this technology reduce and possibly even eliminate previous propinquity and sequential constraints in the human mating market, it increases the opportunity set (for available choice) for potential mates. This
23 increased pool of potential mates also means greater opportunity for selection of partners with lower, similar or even higher levels of certain characteristics, traits or endowments. Education is also an interesting factor as it is commonly used in human mating behaviour as a proxy for resources and future provision helping to gain reproductive or (economic) advantages (Buss,
1985). The online dating environment allows to observe whether individuals select lower, similar, or higher levels of education than themselves providing insights how the selection changes for individual differences (e.g., related to gender or education) across age. Thus, to understand how technology is impacting or facilitating mate choice decisions based on education, we analyse the mating behaviour (rather than mere preferences) of over 41,000 members of the Australian online dating web site RSVP1 and their 219,013 contact decisions across the four months of January to April 2016.
Background
Because choosing a mate can be one of the largest psychological and economic decision humans can make, social science’s extensive explorations of mate choice behaviour represent a broad range of disciplines, including sociology, economics, (evolutionary) psychology and reproductive medicine (Buss, 1985). All these disciplines however, no matter their differences, uniformly acknowledge one phenomenon: positive assortative mating behaviour (homogamy) among humans. Such homogamy is essentially generated by two preference sets in the mating market. In the first, both males and females prefer partners with characteristics, traits or endowments that are symmetrical to their own and so choose each other. In the second, both males and females prefer a particular characteristic, trait or endowment (Schwartz, 2013) – for example, wealth, education or career success – and so choose a similarly wealthy, educated or successful mate. In either case, whether through
1 See https://www.rsvp.com.au/
24 matching or competition, individuals select partners that are more alike than would be the case based on random choice alone.
One advantage of positive assortment is its evolutionary payoff: it ensures the gene transmission optimization that comes from mating with those who share common genes
(Thiessen & Gregg, 1980) while also “increas[ing] the degree to which parents share genes with offspring” (Thiessen et al., 1997, p. 162), which elevates fitness. Symmetrical preferences may thus stem not only from risk minimization but also from an innate recognition mechanism, such as sexual imprinting (Bereczekei et al., 2004). The benefits of positive assortment, however, may go beyond the biological: at a micro level, it can mean increased socio-economic and productivity gains in both the short and long term (Rushton
1988). For example, education-based assortment can confer a range of benefits from improved lifetime health and well-being to increased wages and access to healthcare. It can also bring about greater economic understanding and increased gender equity within a marriage (Shafer, 2013). Assortative mating may be a way to optimise mate choice by selecting a partner with a certain degree of symmetry (Bateson, 1983). This may foster altruistic behaviour inside the family unit, increase marital stability and may even help realise greater fecundity (Rushton, 1988; 1989; Penton-Voak et al., 1999). As such, humans (and other animals) may in fact be able “to detect genetic similarity between themselves and others” (Russell et al., 1985, p.183) leading to a preference for stronger similarity on (highly heritable traits and or) factors like intelligence or education.
As female participation in the labour force and tertiary education increasingly comes to resemble male participation, gender equity between men and women could lead to increasingly symmetrical preferences based on education (Mare, 1991). Because education can be a proxy for resources and their ongoing provision, from a sexual selection perspective,
25 there should be more competition for men and women at the top of the education spectrum
(Rose, 2005). This should normatively further encourage positive educational assortment.
The task of finding a mate has always had opportunity costs because individuals have certain preferences, face constrained supplies and also compete with others to maximize their mate choice (Schwartz, 2013). For example, the traditional geographic constraints of local neighbourhoods, cities and towns meant that the available supply of potential mates was finite, generating a choice constraint in the mating market of the type generally labelled propinquity (Vandenburg, 1972). Yet even in local populations, assortative mating patterns at the phenotypic level always had benefits because individuals could “avoid the costs of leaving the immediate environment to mate” (Penton-Voak et al., 1999, p. 105). The arrival of the
Internet however, has expanded available choices to the point of a quasi-unbounded human mating market. On line, the click of a button results in a myriad of potential mates, thereby reducing potential search costs to virtually zero.
Even though available choices are notionally infinite, socio-economic (and particularly educational) barriers remain. The resulting distance, whether economic or social, means higher inequality, which may in fact increase education or income-based homogamy
(Schwartz, 2013) across the socio-economic distribution. Normatively, these increases in variance may facilitate or encourage males and females to favour the convenience of individuals at each socio-economic equity extreme (Buss, 1985). Such a preference may then propel the remaining individuals with less extreme characteristics (e.g., average educational levels) to pair up based on similarity (Sloman & Sloman, 1988). Generational increases in positive educational homogamy could thus increase inequality and constrain socio-economic opportunity and achievement for offspring (Mare, 1991). Theoretically then greater educational homogamy should increase economic inequality (Schwartz, 2013) because of the education-earnings link. Research shows that residents of less educationally favourable
26 marriage markets are more likely than those in more highly educated markets to marry hypogamously based on education, with the chance of women doing so increasing with age
(Lewis & Oppenheimer, 2000). Research into whether the Internet is facilitating or accentuating changes in educational assortment behaviour and its impact on social equity is still in its infancy.
In addition, as the Internet increases in popularity as a mating tool it may be crowding out more historical social intermediaries like work, school and local neighbourhoods
(Rosenfeld & Thomas, 2012). This is not to say that the geographic proximity of potential partners has become redundant but rather that online searches for mates generally cover a much greater geographic area than the small radius of local neighbourhoods (Rosenfeld &
Thomas, 2012). As this greater coverage translates into a significant increase in available choices, it may also change the structure of mate searches. In particular, new dating technology like the Internet may actually increase the possibility of homogamous selections
(Schwartz, 2013), and in some online dating segments – particularly those facing thin dating markets – it may encourage the convergence of particular characteristics.
This research seeks to identify the factors that influence internet educational assortment behaviour, thereby expanding behavioural science’s understanding of mate choice psychology and decision-making in large scale settings. In particular, by exploring deviations from positive educational assortment, it aims to distinguish the factors influencing education based mate choice decisions in the 2016 online dating market.
Method
Participant pool
27
Our data set, collected from the Australian online dating web site RSVP, encompasses
41,936 unique online dating individuals (2016), who, as part of their web site membership, have provided a wealth of personal details, including height, hair colour, eye colour, body type, sexuality, marital status, ethnicity, religious views, political affiliations, personality type and offspring2. As on most online dating web sites, RSVP users also have the option to provide their ideal preference for each of the characteristics for which they are searching or would prefer in an ideal partner. For the purposes of this study, the participant set is limited to online daters who self-identified as heterosexual (99.60% of the original sample), yielding a final sample with an age range of 18 to 80 years (M = 48.33; SD = 11.25) in which 77.80% of all participants are male.
The data set also includes information on the potential mates contacted by each individual during the four-month investigative period (January 1 to April 26, 2016), which totalled 219,013 contacts with 57,670 separate individuals. Table 1 shows the distribution of educational attainment for the total sample (both participants and their contacts).
4
3
2 Percent
1
0 20 40 60 80 Age of participant
2 Body type, education level and personality type are all measured on a five-point scale (lightest to heaviest, high school to post graduate study, and very private to very social, respectively).
28
Fig. 1. Age distribution (N = 27,695)
Table 1.
Descriptive statistics on participant and contact’s educational background
Participant Participant Freq. Percent /contact Freq. Percent
High school 5,991 14.29% High school 6,533 11.33% Some college 5,387 12.85% Some college 7,039 12.21% Diploma 7,475 17.82% Diploma 10,587 18.36% Degree 10,800 25.75% Degree 16,305 28.27% Post graduate 7,770 18.53% Post graduate 10,429 18.08% Not specified 4,513 10.76% Not specified 6,777 11.75% Total 41,936 100.00% Total 57,670 100.00%
Data Collection
All data were collated by RSVP and provided to the researchers as a single data set.
As collection of member profile demographics and contact behaviour for research purposes are covered under RSVP (FairFax media) Privacy Policy3, all participants have provided informed consent. All demographics collected and used as part of the study (eg. age, sex, body type, etc) was information that was visually available to all RSVP members in each individual participant’s online dating profile.
Empirical model
Our multivariate analysis employs three types of regression modelling: ordinary least squares (OLS), probit and negative binomial. Table 2 reports the regression results using the
OLS and probit models, which treat positive assortment as a binary dependent variable
3 Fairfax Media Privacy Policy – All personal information collected may be used for: “administrative, marketing (including direct marketing), promotional, planning, product/service development, quality control and research purposes, or those of our contractors or external service providers”
29 clustered by each individual online participant. The independent factors chosen for our multivariate analysis are all commonly identified determinants of mate choice across a wide range of behavioural science research (Buss, 1985; Buss & Barnes, 1986; Botwin et al., 1997;
Hitsch et al., 2010a). All of the relevant factors used are also immediately observable by online daters when viewing and deciding on potential mates dating profiles. As such participants’ decision to contact a specific member of the dating website community is done in the complete knowledge of similarity or difference within each particular variable stated.
30
20 Percent
10
0 -4 -3 -2 -1 0 1 2 3 4 < Hypergamy - Hypogamy >
Fig. 2. Educational hypergamy to hypogamy (N = 43,147).
The independent variables are online dater demographics, including age, sex (dummy for male), height, body type, personality, education and marital status (dummy for single), as well as dummies for no children, wanting (more) children, ethnicity, political views and religious ideology4. Because each online dating contact initiation is independent of all others, we also employ a negative binomial regression (using a count-data model) to more accurately explore discrete probability in online dating behaviour.
4 For a description, see Table 2.
30
Deviation from positive educational assortment can be referred to as educational hypergamy (choosing a mate with a higher level of education relative to your own) or educational hypogamy (choosing a mate with a lower level of education relative to your own).
To measure this deviation from positive assortment (see Table 2) and derive a new hypergamy/hypogamy variable, we subtract the value of the contact’s continuous education variable from the value of the participant’s to provide a positive or negative measure of difference from positive assortment (the zero value). A positive (negative) value (i.e., one greater (less) than zero) means that the individual is contacting another online dater with a lower (higher) level of education than his or her own (hypogamy or hypergamy, respectively).
We graphically depict this variable in Figure 2, with the zero value representing positive educational assortment by the online daters. Then, in Table 3, we identify the factors at play when online daters contact others with lower education levels (educational hypogamy) by using only the positive values (specifications 5 & 7) of the (dependent) hypergamy/hypogamy variable in OLS and negative binomial regressions.
We similarly identify the relevant factors for online daters contacting those with higher educational levels (educational hypergamy) using the hypergamy/hypogamy variable’s negative values (specifications 6 & 8, Table 3) in OLS and negative binomial regressions
(Nbreg)5 that are clustered by unique individual online user. For all OLS results, we show the relative strength of the variables by reporting robust beta-standardized coefficients, and for all the probit models, we identify the quantitative effects by reporting marginal effects.
Results
Specification (1) (first column of Table 2) uses OLS and the binary dependent variable of participants who contacted another online dater with the same educational level
5 Because negative binomial regression requires positive values, all values below zero are converted to positives.
31
(educational homogamy). Here, positive educational assortment takes the value of one, while all other combinations of participant and contact educational level are equal to zero. In all models, dummy variables control for the categorical factors of ethnicity, political views and religious ideology. Specification (2) then extends this specification by controlling for the participant’s political views and religious beliefs.
Overall, the analyses indicate that higher educational levels are positively correlated with contacting online daters with symmetrical educational levels. Specification (4), for example, reveals that a one unit increase in education raises positive assortment by 5.3%. The beta coefficients also show that education has the largest standardized coefficient, with a one standard deviation increase raising the standard deviations in positive assortment by 0.152
(specification (2)). The results also indicate that those with no children are more likely to contact online daters with a symmetrical educational level, while wanting (more) children is positively correlated with positive assortment but only at a 10% statistical significance in the fuller specifications. The analyses also show that those with heavier body types are less likely to assort based on educational level, although age, sex, height, personality and being single exhibit no statistically significant relation with positive educational assortment in this setting.
32
Table 2. Positive educational assortment (based on OLS & probit regressions) Dependent variable: Educational homogamy (1) (2) (3) (4) OLS OLS Probit Probit Age -5.7e-05 -7.7e-05 -7.1e-05 -1.4e04 (1.4e-04) (1.4e-04) (4.9e-04) (4.9e-04) -0.002 -0.002 -2.1e-05 -4.1e-05 Male -0.004 -0.003 -0.017 -0.014 (0.004) (0.004) (0.013) (0.013) -0.004 -0.003 -0.005 -0.042 Height -2.2e-04 -2.5e-0.4 -0.001 -0.001 (2.2e-04) (1.9e-04) (0.001) (0.001) -0.004 -0.005 -2.4e-04 -2.8e-04 Body type -0.004** -0.003** -0.010** -0.010** (0.001) (0.001) (0.005) (0.005) -0.007 -0.007 -0.003 -0.003 Personality 0.002 0.002 0.007 0.009 (0.002) (0.002) (0.006) (0.006) 0.003 0.004 0.002 0.003 Education 0.051*** 0.049*** 0.175*** 0.171*** (0.001) (0.001) (0.003) (0.004) 0.162 0.157 0.053 0.052 Marital status: single 3.3e-04 3.7e-04 0.002 0.002 (0.004) (0.004) (0.013) (0.013) -2.3e-04 0.000 0.001 -5.5e-04 No children 0.008*** 0.007** 0.024** 0.021** (0.003) (0.003) (0.010) (0.010) 0.009 0.008 0.007 0.006 Wanting (more) children 0.006 0.007* 0.019 0.023* (0.004) (0.003) (0.014) (0.014) 0.006 0.016 0.006 0.007 Ethnicity Yes Yes Yes Yes
Political views No Yes No Yes
Religion No Yes No Yes
N 187,713 187,713 187,713 187,713 N Clusters 35,135 35,135 35,135 35,135 Prob > F / Prob > χ2 0.0000 0.0000 0.0000 0.0000 Pseudo R2 0.0280 0.0284 0.0268 0.0271 Notes: Standard errors are in parentheses; beta coefficients (OLS regressions) and marginal effects (probit regressions) are in italics. The beta coefficients were derived by running separate regressions without clustering. Political views (dummy) = left wing, right wing, no strong beliefs or swing voter; religion (dummy) = agnostic, atheist, Anglican, Catholic, born again
33
Christian, other Christian, New Age, Islamic, Jewish, Buddhist, Hindu, Sikh, or other; ethnicity (dummy) = Northern European, Southern European, Eastern European, Hispanic/Latino, other Caucasian, Black African, Middle Eastern, Asian, Indian, Aboriginal, Other Islander, mixed race, other ethnicity. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels, respectively.
Table 3. Educational hypogamy and hypergamy (based on OLS and negative binomial regressions)
Dependent variable: (5) (6) (7) (8) Educational Hypogamy Hypergamy Hypogamy Hypergamy hypogamy & hypergamy OLS OLS Nbreg Nbreg Age -0.002*** 0.003*** -0.001*** 0.002*** (4.1e-0.4) (0.001) (0.000) (3.2e-0.4) -0.029 0.037 -0.002 0.003 Male -0.007 -0.102*** -0.006 -0.061*** (0.011) (0.014) (0.006) (0.008) -0.003 -0.053 -0.010 -0.101 Height -1.2e-04 -0.003*** 6.4e-05 -0.002*** (0.001) (0.001) (0.000) (4.2e-0.4) -0.001 -0.034 0.000 -0.003 Body type -0.021*** 0.053*** -0.010*** 0.032*** (0.004) (0.005) (0.002) (0.003) -0.019 0.052 -0.018 0.052 Personality 0.008 0.001 0.004 0.001 (0.005) (0.006) (0.003) (0.004) 0.006 0.001 0.007 0.002 Education -0.525*** 0.221*** -0.307*** 0.140*** (0.004) (0.004) (0.002) (0.003) -0.630 0.232 -0.526 0.229 Marital status: single 0.005 0.012 4.8e-04 0.007 (0.008) (0.013) (0.006) (0.008) 0.002 0.004 0.001 0.011 No children 0.057*** -0.033*** 0.031*** -0.020*** (0.008) (0.011) (0.007) (0.007) 0.031 -0.019 0.053 -0.033 Want (more) children 0.039*** -0.067*** 0.020*** -0.042*** (0.012) (0.015) (0.006) (0.009) 0.016 -0.030 0.034 -0.067 Ethnicity Yes Yes Yes Yes
Political views Yes Yes Yes Yes
Religion Yes Yes Yes Yes
N 68,682 54,113 68,682 54,113 N Clusters 20,116 17,949 20,116 17,949 Prob > F 0.0000 0.0000 0.0000 0.0000 R-squared 0.3857 0.0586 0.0641 0.0096 Notes: Standard errors are in parentheses; beta coefficients (OLS regressions) and marginal effects (probit regressions) are in italics. The beta coefficients were derived by running separate regressions without clustering. Political views (dummy) = left wing, right wing, no strong beliefs or swing voter; religion (dummy) = agnostic, atheist, Anglican, Catholic, born again
34
Christian, other Christian, New Age, Islamic, Jewish, Buddhist, Hindu, Sikh, or other; ethnicity (dummy) = Northern European, Southern European, Eastern European, Hispanic/Latino, other Caucasian, Black African, Middle Eastern, Asian, Indian, Aboriginal, Other Islander, mixed race, other ethnicity. *, ** and *** denote statistical significance at the 10%, 5% and 1% levels, respectively.
To further explore the effect of the age-education interaction on positive educational assortment, we graph the average marginal effects in Figure 3 (based on specification (2)), which shows that the higher the educational level online daters (in particular younger ones), the more likely they are to contact others with homogenous educational levels. However, the slope for individuals with higher educational levels is negative and steeper (see the three highest educational groups) indicating that the older cohorts care less about educational homogamy while cohorts with lower educational levels care more (see positive slope) as they get older. Thus, older online daters become more similar in regards to how much they care about homogamy.
High School Some College Diploma Degree Post Graduate
.4
.3
.2 Linear Prediction Linear .1
0 18 80 Age of participant
Fig. 3. Predicted margins for assortative mating, with a focus on age slopes as a function of education (linear prediction).
35
Table 3 reports the results for the four specifications (two OLS and two Nbreg) that explore the factors which influence online daters’ educational hypogamy ((5) and (7)) and educational hypergamy decisions ((6) and (7)). In these analyses, those who are older, have heavier body types, or higher educational levels are all less (more) likely to contact those with lower
(higher) educational levels than themselves. For example, a one unit increase in education decreases hypogamy by 52.5% (see (5)) while increasing hypergamy by 22.1% (see (7)).
Conversely, online daters who have no children and those who want more children are more
(less) likely to hypogamously (hypergamously) sort based on education. Yet again, sex, height, personality and being single have no statistically significant coefficients. Overall, therefore, education has the largest relative importance, as indicated by the largest standardized coefficient.
To further explore decisions to deviate from positive educational assortment, we identify the interaction effects between age and educational hypogamy or hypergamy by gender (see Figures 4 and 5). We find that younger male online daters are less likely than younger females to contact daters with higher educational levels (Figure 4). On the other hand, females across all age groups show very little change in their willingness to contact males with higher educational levels (a linear prediction of 1.7 to 1.75) and are in fact more likely to mate hypergamously based on education. Nevertheless, older online daters do exhibit similar levels of sex-based differences in hypergamous contact with other potential mates online.
In addition, as Figure 5 shows, younger males are more likely than younger females to use educational hypogamy as a strategy when contacting online daters of the other sex. On the other hand, online daters over 50 show the opposite trend: older females are more likely than older males to contact individuals with educational levels lower than their own.
36
Female Male
1.75
1.7
1.65
1.6 Linear Prediction
1.55
1.5 18 80 Age of participant
Fig. 4. Predicted margins for educational hypergamy, with a focus on age slopes by gender (linear prediction).
Female Male
1.9
1.85
1.8 Linear Prediction Linear 1.75
1.7 18 80 Age of participant
37
Fig. 5. Predicted margins for educational hypogamy, with a focus on age slopes by gender (linear prediction).
Discussion
Our analysis of the factors influencing homogamous, hypogamous and hypergamous decisions to contact potential mates online indicates that, consistent with previous research
(Buss, 1985), more highly educated online daters are more likely to contact others with the same educational level. The interaction effects further demonstrate that positive assortment becomes similar among older online daters across all levels of educational achievement, while younger more educated online daters are more likely to mate homogamously. The results for hypogamy similarly show that older more educated online daters are less likely than other participants to contact those with lower educational levels.
On the other hand, sex differentiating these results in our analysis of interaction effects reveals that younger males are more likely than younger females to contact someone with lower educational levels, which confirms previous evidence of males not being averse to contacting less educated females (Skopek et al., 2008). Interestingly, older women are also more likely to contact less educated men, suggesting a sex difference intersection of male and female educational hypogamy preference at approximately 50 years of age. One possible explanation offered by qualitative research on older online dater preferences is that males favour committed relationships, whereas females may “desire companionship without demanding caring roles” (McWilliams & Barrett, 2014, p. 411). If females do in fact become less selective with age relative to males, then their participation in the online dating market is
38 likely to focus more on ongoing interaction rather than securing an exclusive relationship with a potential male companion.
Both the educational hypergamy and the interaction effects analyses suggest that older online daters and women are more likely to contact a potential mate with a higher educational level, although the sex difference diminishes considerably among older RSVP members.
From an evolutionary perspective, the concept of “female choice” (Buss & Barnes, 1986, p.
559) may provide some insight into women’s (vs. men’s) relative selectivity for those with higher educational levels. That is, the opportunity cost of participating in the reproductive market is inherently greater for women (Trivers, 1972), so whether seeking more offspring or not, females in the online dating market may still exhibit evolved adaptive preference for resources (higher education) and proxies for the ability or willingness to provide them.
Naturally then, as in traditional settings, female online daters initiate contact and reply more selectively than males (Fiore et al., 2010). Online dating research on middle age to older online daters has also shown that female selection criteria centre more on a male’s abilities
(McWilliams & Barrett, 2014), which could explain males’ comparatively lower preference for educational hypergamy and increased preference for educational hypogamy in potential female mates. Given that male education historically exceeds that of females (particularly among older participants), hypogamy implies a larger potential market of females seeking more educated males from which male daters can choose.
Positive assortment is also more likely among online dating participants with no children and those who want more children, although these are also more (less) likely to contact less (more) educated others. These results align with research findings among online daters of significant homophily for life course preferences such as wanting children (Fiore &
Donath, 2005). Previous research has also shown significant homophily in self-reported physical build for online daters (Fiore & Donath, 2005). Our analysis of the body-type
39 education relation, shows those with heavier body types are less likely to positively assort, less likely to mate hypogamously and more likely to contact more highly educated
(hypergamy) potential mates.
The authors also acknowledge several limitations to the current study. This study is limited by its use of a sample derived from only one dating web site, meaning that the research findings, although based on an extremely large and diverse sample, may in part reflect the platform provided by the particular web site and its participants. The time of year may also have impacted behaviour given that different seasons or months can translate into a selection effect. In addition, even though humans have an amazing ability to collate and analyse information through interaction and observation (Wilson & Dugatkin, 1997) because the Internet can theoretically provide infinite choice in a non-sequential setting, some online daters may have experienced cognitive overload or similarly negative cognitive outcomes during their cyber search for a potential mate. Research has also demonstrated that, for both sexes, increased choice can mean reduced utility and increased difficulty in decision making
(Lenton et al., 2008). Whether or not these difficulties facilitate or accentuate homogamous, hypogamous or hypergamous contact behaviour in an online dating setting is less clear and thus warrants further investigation.
A further limitation of the study is information asymmetry and deception by online dating participants. Whether consciously or not the conduit of the Internet certainly allows for exaggeration in self-representation. However, this does not mean that deception is the most common or successful strategy employed in the cyber dating world. As the payoff for online dating is how well the face to face meeting goes, participants are constrained in their ability to over or understate particular features. In fact research has shown that while online daters may provide false information in their profiles, the inaccuracies are limited and often so minute
(eg. height, weight, age) that they would be difficult to identify in any face to face contact
40
(Ellison et al., 2006; Toma et al., 2008; Guadagno et al., 2012). It is for this very reason that wider social science uses such readily available (online) data sources to study human decision making and mate choice behaviour.
Conclusion
Because the burgeoning online dating conduit has significant potential to create a more efficient and successful mechanism for finding and securing both short-term and long- term romantic relationships (Finkel et al., 2012), ever greater numbers of people are choosing to use it to find a partner. Practical implications for understanding human cyber-dating behaviour extend not just too individual relationship formation, but also couples counselling psychology and psychotherapy. It is for these reasons that further research is warranted into the impact of technology in the mating market, and particularly on the psychology employed by humans when using the Internet to maximize their educational preference in a mate.
41
Chapter 3: PREFERENCE VS. CHOICE IN ONLINE DATING
Stephen Whyte & Benno Torgler
Cyberpsychology, Behavior and Social Networking (2017) 20(3), 150-156
Abstract
This study explores factors that influence matches of online dating participants’ stated preference for particular characteristics in a potential partner and compares these with the characteristics of the online daters actually contacted. The nature of online dating facilitates exploration of the differences between stated preference and actual choice by participants, as online daters willingly provide a range of demographics on their ideal partner. Using data from the Australian dating website RSVP, we analyze 219,013 contact decisions. We conduct a multivariate analysis using the number of matched variables between the participants’ stated preference and the characteristics of the individuals contacted. We find that factors such as a person’s age, their education level and a more social personality all increase the number of factors they choose in a potential partner that match their original stated preference. Males (relative to females) appear to match fewer characteristics when contacting potential love interests. Conversely, age interaction effects demonstrate that males in their late 60’s are increasingly more selective (than females) regarding who they contact. An understanding of how technology (the Internet) is impacting human mating patterns and
42
the psychology behind the participants, informs the wider social science of human behavior in large scale decision settings.
Introduction
The Internet has become such a common and widely used form of technology that the online population is increasingly homogenous with the offline population (Valkenburg & Peter,
2007). Since its inception (less than two decades ago) the Internet has facilitated a fundamental shift in how humans are able search and interact with potential partners (Finkel et al., 2012). The cyber world is becoming one of the most widely accepted and popular tools for finding both short and long term love interests (Heino et al., 2010; Dröge & Voirol, 2011)
. It has even partially displaced more traditional avenues such as family and friends, work and the local community as a conduit for first contact with that special someone (Rosenfeld &
Thomas, 2012). Research from Britain shows the third most popular method of finding a date is the Internet (Couch & Liamputtong, 2008). More than one in ten American adults – and
38% of those who identify as “single and looking” for a partner – have used some form of online dating platform (Smith & Duggan, 2013) In the United States alone there is in excess of 40 million unique American dating website users as of 2010 (Sritharan et al., 2010)
At a micro level, people use online dating to search for a partner for a range of reasons. Increased information, convenience and accessibility, age, greater choice, or an inability to meet potential mates via their own social networks (Couch & Liamputtong, 2008;
43
Frost et al., 2008). It is little wonder then, that online dating has become a setting for behavioral science to explore the psychology of human interaction and decision making.
There have been numerous studies using online dating for the observation of behaviour
(Bapna et al., 2013; Hitsch et al., 2010a; Lee et al., 2008; Pizzato et al., 2010; Wu & Chiou,
2009). Online dating is a unique domain, often requiring individuals to provide extensive information on themselves, on their preferences for their ideal partner, and even their resulting mate choice. The contact and interaction history offers a rich source of implicit information
(Lee et al., 2008). From a research perspective, the greatest benefit of online dating data is that it is in individuals’ best interests to provide as much (honest) information as possible.
Minimizing asymmetric information between potential matches reduces transaction and search costs for all market participants.
While online dating provides many individuals an otherwise inaccessible market of new potential mates, questions may arise around such platforms ability to allow participants to exaggerate or lie in user profile formation. On the surface this may seem the most efficient strategy, but the payoff for any online dating interaction is to meet in person or a date. While it may be possible to exaggerate one’s potential to get selected, online self-representation cannot be too exaggerated if the end goal is to have a successful and enjoyable face-to-face date (Toma & Hancock, 2010). It is costly to send deceptive or noisy signals, assuming that the final payoff for all participants is to have a successful first meeting. In fact, research shows that honesty is the best policy, as there is often little or no evidence of strategic behavior in online dating settings (Hitsch et al., 2010a). While online daters may emphasize their positive characteristics (Guadagno et al., 2012) in their advertised dating profiles, inaccuracies are often small (e.g., related to factors such as age, height and weight) perhaps because they are difficult to detect in person (Toma et al., 2008). Ellison, Heino and Gibbs
(2006) go as far as to say that the idea that online daters “frequently, explicitly, and
44 intentionally “lie” online is simplistic and inaccurate”. Moreover, when comparing the reported height of the people in our data set with information from the Australian Bureau of
Statistics we find that the stated height of both men and women is only slightly above the
Australian average (by less than 4 centimeters) which indicates little evidence of misrepresentation.
Such findings are also pertinent irrespective of the sexual (short term) or relationship
(long term) objectives or motivations of the individuals involved. Whether searching for romantic or sex-only relationships, participant’s end goal of face-to-face contact ensures men and women’s online dating information (profiles) are an apt reflection of their actual demographic characteristics. A significant (social and) market mechanism ensuring participants are accurate and honest in their representations. It is for this very reason that social science now extensively uses internet interactions, and particularly online dating, as a rich and credible source of quantitative behavioral data (Hitsch et al., 2010a; Lee & Niederle,
2015).
The domain of online dating, then naturally facilitates exploration of the differences between stated (advertised) preference and actual choices of participants. In line with adaptive decision making theories such as satisficing (Simon, 1972), online dating preference research shows that explicit preferences are weak indicators of the success of user interactions (Frost,
2009). In fact, research into actual choices made by individuals is far more informative as dating participants’ actions may be somewhat contrary to their original stated preference
(Heino et al., 2010). Exploring the progression employed between what humans say they want and what they actually choose is of critical importance for behavioral science. And as wider social science is often the academic narrative for public policy (e.g. health and education expenditure), the benefits of understanding how humans match and choose cannot be
45 understated. Large scale settings such as online human mating scenarios provide a fitting domain for such psychological investigation.
Method
Participants
Our data set consists of 41,936 unique individual online dating members’ “dating profile” information from the Australian dating website, RSVP (Whyte & Torgler 2017). A member’s profile contains a range of demographic self-reported data, such as age, height, eye colour, hair colour, body type, sexuality, marital status, ethnicity, religious views, political affiliations, and number of offspring. Data are also captured on participants’ stated preference regarding the ideal mate or love interest that they are searching for on RSVP. Males make up
77.80% of the sample, with close to one hundred percent (99.60%) of all participants identifying themselves as heterosexual. For the purpose of this study, only heterosexual participants are used in our analysis. Participant age ranges from 18 to 80 years of age. RSVP members mean age being 48.33 years with a SD of 11.25 years. RSVP members mean age is
48.33 years with a SD of 11.25 years with 50% of all participants being between the ages of
39 and 56 years old. Comparing our sample with statistics from the wider Australian population taken from the Australian Bureau of Statistics (ABS, 2016) indicates that our cohort is on average 11 years older. In relation to the sex ratio of participants our sample mirrors previous research with males being over represented (Hitsch et al., 2010a). This may reflect the fact that men tend to visit online dating websites more often compared to women
(Heino et al., 2010) and are more likely to contact women whose profiles they viewed
(Valkenburg & Peter, 2007).
Procedures
46
The data set also contains actual decision (choice) information on whom the 41,936 individuals contacted via the dating website between the 1st of January 2016 and the 26th April
2016. There were a total of 219,013 contacts made, to 57,670 unique individuals. The number of matching variables between the participants’ stated preference (what they want) and the characteristics of the individuals they actually contacted are used as our dependent variable in the multivariate analysis. The seven matching factors are: hair colour, eye colour, body type, education level, personality type, political view, and religious affiliation6.
As Figure 6 shows, more than one in every three contact decisions by RSVP members was to make contact with a potential partner that matched zero of the seven preferences they had originally stated. And more than 65% of the total 219,013 contacts initiated had one or less variables matching their stated preference.
40
30
20 Percent
10
0 02468 Number of matching characteristics
Fig. 6. Percentage of participants contacted matching up to 7 characteristics
6 Body type, education level and personality type all being five point scales (lightest to heaviest, high school to post graduate study, and very private to very social, respectively).
47
Results
To gain an overview of determinants impacting the psychology of mate preference matching actual mate choice, we perform an ordinary least squares (OLS) regression analysis, clustering by individual. In specification (1) we begin with the physical and aesthetic features of participant age, a dummy variable for sex difference, height, body type, and ethnicity. We find that older participants contact potential partners that match more of their preferences, while male and taller individuals match less. Body type shows no influence. In specification
(2), we now introduce personality based characteristics (a self-identified personality measure of how “social” the participant regards themselves). Due to missing values in our data set (not all online daters state all of their preferences/personal characteristics) our sample size then falls from 213,843 observations to 187,713 observations. People with more social personalities are more inclined to contact those who match more of their original preferences.
In specification (3) we introduce and control for circumstantial factors like current level of education, current relationship status, offspring from previous relationships, and future intention for more children. All of these new factors appeared not to matter for actual matching choices that align stated preferences. Finally, in specification (4) we include a further set of controls (dummies for political views and religious ideology) as a further robustness check. The key results for age, males, and personality remain robust. We present also beta/standardized coefficients to explore the relative importance of the variables, and put all the factors into a common metric. In all equations age has the largest standardized coefficient. For example, in specification (4) we can see that an increase of one standard deviation in age is correlated with an increase of 0.096 standard deviation in matching.
48
Taking into account that our dependent variable is a count variable with a high proportion of zeros (see Fig. 6), we conduct a further robustness check by generating results using the zero-inflated model (see Table 4). To discriminate between the negative binomial and the zero-inflated model we apply a Vuong’s test. The test statistics (z between 20.09 and
29.52) indicate that the zero-inflated model should be used. We find consistent results. The coefficients of the variables age, male dummy, and personality remain statistically significant.
However, height is now also statistically significant in all four specifications.
Table 4. OLS analyzing matches of stated preference and actual choice
Dependent Variable: (1) (2) (3) (4) Number of matches Age 0.018*** 0.018*** 0.018*** 0.013*** (0.001) (0.001) (0.002) (0.002) 0.138 0.133 0.130 0.096 Male dummy -0.137*** -0.136*** -0.133*** -0.158*** (0.041) (0.042) (0.046) (0.043) -0.038 -0.038 -0.036 -0.043 Height -0.004* -0.004* -0.004* -0.003 (-0.002) (0.002) (0.002) (0.002) -0.021 -0.022 -0.025 -0.016 Body type -0.021 -0.019 -0.018 -0.028 (0.017) (0.017) (0.018) (0.018) -0.011 -0.010 -0.010 -0.015 Personality 0.082*** 0.086*** 0.088*** (0.019) (0.021) (0.020) 0.039 0.040 0.041 Education 0.012 0.021 (0.011) (0.011) 0.011 0.019 Marital Status: Single 0.123 0.111 (0.076) (0.074) 0.025 0.022 No children dummy 0.004 0.002 (0.037) (0.036) 0.001 0.001 Want more children 0.014 -0.038
49
(0.042) (0.042) -0.003 -0.010 Ethnicity dummy Yes Yes Yes Yes
Political dummy No No No Yes
Religion dummy No No No Yes
N 213,843 204,986 187,713 187,713 N Clusters 40,601 38,765 35,135 35,135 Prob > F 0.0000 0.0000 0.0000 0.0000 R-squared 0.0616 0.0608 0.0575 0.0792 Notes: Ordinary Least Squares Regression standard errors are presented in parentheses. Beta coefficients are presented in italics. Separate regressions were run without clustering to produce beta coefficients. Seven characteristics are: Hair colour, eye colour, body type, education, personality, political view, religious affiliation. Categories for dummy controls are, Political: Left wing, right wing, no strong beliefs, swinging voter. Religion: Agnostic, Anglican, atheist, born again Christian, Buddhist, Catholic, Hindu, Islamic, Jewish, New Age, other, other Christian, Sikh. Ethnicity: North European, South European, East European, Hispanic Latino, Other Caucasian, Black African, Middle Eastern, Asian, Indian, Aboriginal, Other Islander, Mixed Race, Other Ethnicity. *, ** and *** represent statistical significance at 10, 5 and 1% levels, respectively.
Table 5. Negative binomial model analyzing matches of stated preference and actual choice
Dependent Variable: (1) (2) (3) (4) Number of matches Age 0.011*** 0.011*** 0.009*** 0.006*** (0.001) (0.001) (0.001) (0.001) Male dummy -0.044 -0.057 -0.064* -0.088*** (0.034) (0.035) (0.034) (0.031) Height -0.005*** -0.005** -0.004** -0.003* (0.002) (0.002) (0.004) (0.002) Body type -0.015 -0.014 -0.008 -0.007 (0.015) (0.015) (0.008) (0.013) Personality 0.087*** 0.078*** 0.074*** (0.023) (0.021) (0.020) Education -0.007 -0.002 (-0.007) (0.009) Marital Status: Single 0.071 0.069 (0.050) (0.047) No children dummy -0.027 -0.027 (0.028) (0.027) Want more children -0.064 -0.065 (0.039) (0.041) Ethnicity dummy Yes Yes Yes Yes
Political dummy No No No Yes
50
Religion dummy No No No Yes
Ethnicity dummy No No No Yes
N 213,843 204,986 187,713 187,713 Non zero N 137,905 134,556 126,703 126.073 N Clusters 40,601 38,765 61,640 61,640 Prob > chi2 0.0000 0.0000 0.0000 0.0000 Vuong Test z=20.27 z=20.09 z=23.62 z=29.52 Notes: Zero-Inflated Negative Binomial Regression robust standard errors are presented in parentheses. Seven characteristics are: Hair colour, eye colour, body type, education, personality, political view, religious affiliation. Categories for dummy controls are, Political: Left wing, right wing, no strong beliefs, swinging voter. Religion: Agnostic, Anglican, atheist, born again Christian, Buddhist, Catholic, Hindu, Islamic, Jewish, New Age, other, other Christian, Sikh. Ethnicity: North European, South European, East European, Hispanic Latino, Other Caucasian, Black African, Middle Eastern, Asian, Indian, Aboriginal, Other Islander, Mixed Race, Other Ethnicity. *, ** and *** represent statistical significance at 10, 5 and 1% levels, respectively.
Due to the strongest effect of age on matching we take a closer look at whether age interacts with other factors by applying specification (4) in Table 4. Thus, we fit a separate slope for the relationship between age and matching for male and female. The age slope for male and female is significantly different (t=3.01). The hypothetical predicted values from such a model are shown in Figure 7. The slopes visually depict younger men matching fewer characteristics than women from the age of 18 up until their late 60’s. After this males become more selective with whom they contact, compared to their female counterparts, matching more of their stated preference when contacting potential mates.
The same modelling process is repeated to explore interaction effects between age and education (Figure 8). The slopes represent visually those online daters who have greater levels of education matching less characteristics of their preference in who they contact for younger ones, relative to those with less education. Inversely those with higher levels of education and over the age of 40 are more selective (matching more characteristics to their stated preference) relative to those with less education.
51
1.8 1.6 1.4 1.2 1
18 80 Age of Participant
Female Male
Fig. 7. Average marginal effect: Age and Gender
2 1.8 1.6 1.4 1.2 1
18 80 Age of Participant
High School Some College Diploma Degree Post Graduate
Fig. 8. Average Marginal effect: Age and Education
52
Discussion
This study explores factors that influence matches between online dater’s stated preference for particular characteristics in a potential partner and the characteristics of the online daters they actually contact. Factors such as a person’s age, their height, and their personality are all positively correlated with the number of factors in a chosen potential partner that match their originally advertised preference. Males (relative to females) appear to match fewer characteristics when contacting potential love interests online, at least until they reach their late 60’s. Also, relative to those with less education, more educated younger online daters match fewer characteristics when contacting a potential mate. Conversely older online daters with higher levels of education match more of their preferences in the people they contact.
The finding that older participants match more preferred characteristics in those they contact is interesting. While online dating has been extensively studied across behavioural science for more than a decade, (arguably) less attention has been given to the psychology and behavior of older online daters. Millennials and university students are an exhaustively used subject pool in both online and laboratory research, as more mature samples are often much harder and costly to observe. But as the internet facilitates increased participation across generations, more and more non-millennials engage in the conduit for mating that it provides.
Of the current research on mature daters, evidence shows that men are more likely to date
(Watson & Stelle, 2011) and those participating in the dating market have higher levels of education, having accrued more assets, and self-identify as having more social connectedness
(Brown & Shinohara, 2013). Older daters also appear to set higher standards for potential mates’ educational levels (Buunk et al., 2002), a finding that is not surprising considering educational attainment naturally accrues with age. Moreover, research in speed dating settings has also shown that men and women do not necessarily adjust (with age) how selective they
53 are (Kurzban & Weeden, 2005). Older online daters may not necessarily be more selective than their younger counterparts; rather they just contact less of those who do not meet their preferred characteristics in a mate.
Evolutionary psychology has put forward motivations for strategic sex difference in human mating (Trivers, 1972). Our results indicate that males contact more online daters with less characteristics that match their stated preference, building on the idea of “female choice”, and that men are less selective in who they choose to mate with (Buss & Barnes, 1986).
Whether men realize the evolutionary payoff of more offspring from this strategy in an online dating market is unclear. While the technology of online dating has changed the medium of the dating market, women still appear to be more selective than males with whom they contact online (Alterovitz & Mendelsohn, 2009). It is also interesting that males in their sixties surpass their female equivalents and match more of their advertised preference in who they contact. While the composition of the online dating market may reflect disproportionate sex difference across different age groups, such a significant change warrants further research.
As Australian males have shorter life expectancy than females (Mathers et al., 2003), one hypothesis may be that older males may in fact be looking for a cooperative care arrangement as well as a love interest. Thus, men in their late 60s may prioritise their preferences accordingly, giving greater weighting to women who have the potential to also fill the role of caregiver (McIntosh et al., 2011). Or alternatively, for the older cohort significant mating pool sex difference may just mean females are less selective because they have less potential mates to choose from.
That more outgoing or social personality types realise more matches in preference and choice is also a notable finding. On the surface we would expect more extroverted people to cast a wider net of both attention and interaction in the mating market. In any market, the ability to sell or advertise a good or service can facilitate increased consumption. Being a
54
“good seller” may translate into less necessity to contact as many potential mates. Ofcourse searching for and matching characteristics in suitable partners (to your preference) comes at a higher opportunity cost. From an evolutionary sense, it would also significantly reduce any pool of potential mating options, as fewer participants could match all stated preferences.
More social and extroverted individuals may in fact believe their personality type is more appealing (Hall et al., 2010). Having a more social personality type may also grant participants the confidence to be more selective in who they contact in the cyber dating world.
Conclusion
Disclosure of partner preferences is a widely offered and commonly used option for participants of online dating websites. The core premise is that it allows participants to advertise (signal) to potential love interests who may be endowed with or in possession of said characteristics and traits. The efficiency of this mechanism in eliciting an understanding of actual choice and decision behavior is not as clear. Speed dating research shows that advertised preference can be an indication of willingness to participate or attend, but may provide very little predictive capacity with respect to actual choice. Economics alludes to this fact with the idea that individuals generally tend to overstate their valuations (Murphy et al.,
2005), and advertising one’s ideal partner preferences online may just be another microeconomic example. Of course as online dating’s natural progression is towards face to face dating, preferences may just be a “wish list” of individual demographics, rather than a rigid mate choice criterion. Meeting that special someone (with enough of the characteristics we prefer) may be the optimal and most flexible strategy (Wiederhold, 2015).
Certainly, both men and women know that they are in a mating market, and respond or adjust to market forces accordingly (Pawlowski & Dunbar, 1999). Humans are conscious of the changing conditions of online markets, and how it differs from other sequential settings
55 like local night spots, school, work, or social gatherings. As the psychology employed by humans when choosing a partner can certainly be environmentally sensitive, online dating may trigger changes in underlying preferences and decision behavior (Kurzban & Weeden,
2007). A disproportionate amount of choice relative to other everyday consumption decisions may impact or even change our original stated preference. Rather than seeking the specifically highest utility payoff available in a mate, human beings may make decisions that are optimal in relation to the mental, resource, age, and environmental constraints they are faced with when searching and choosing (Simon, 1972).
In this fast paced world, and with the myriad of options the internet now provides, time spent searching and exploring all available potential partners is extremely costly. More than ever before, our cognitive bounds may in fact be aiding us in our adaptive decision making (Todd, 2007). This means searching and evaluating through suitable options until one meets an acceptable threshold, rather than matching a single or set of specific idealized preferences. The online dating decision making process and our results demonstrating a lack of matching preference with actual choice reflects a wider willingness to contact any and all potential mating options available. As the Internet and online dating exponentially grow, further research is warranted in the psychology and decision processes employed by online daters and the relationship between stated preference and choice.
56
Chapter 4: DO WOMEN KNOW WHAT THEY WANT? SEX DIFFERENCES IN ONLINE DATERS EDUCATIONAL PREFERENCES
Stephen Whyte, Ho Fai Chan & Benno Torgler
Psychological Science (2018) forthcoming
Abstract
57
Using a unique cross-sectional data set of dating site members’ educational preferences for potential mates (N = 41,936), we show that females are more likely than males to stipulate educational preferences at all ages. When members indifferent to educational level are excluded, however, the specificity of male and female preferences does differ across age groups. That is, whereas females express more refined educational preferences during their years of maximum fertility, their demand specificity decreases with age. Male specificity, in contrast, remains stable until the 40s, when it surpasses that of post-reproductive females and then increases as males reach their peak career earnings potential. The greater female investment in offspring also translates into more selective preferences for potential mates when the offspring is still living at home.
Introduction
Sexual selection decisions are at the core of evolutionary psychology human mating strategies, including parental investment, whose levels tend to differ between males and females (Trivers, 1972). Because “[t]hose who hold valuable resources do not give them away cheaply or unselectively” (Buss, 1994, p. 33), this parental investment asymmetry from different minimum physiological obligations (fertilization, gestation and lactation), sex cell size and mobility implies that men, having lower sex investment, will be less choosy in mate preferences (Schmitt, 2005). Evolution thus favours women who are highly selective about their males, focusing on quality over quantity. These males, however, vary significantly across multiple factors, which forces a trade-off that over millennia has trained females to
58 identify adaptively valuable characteristics while recognizing their own wants and preferences for particular traits. This act of choosing, therefore, is a difficult task in which the decision maker must not only weigh the immediate attractiveness of options but also anticipate the consequences of choice (Schwartz, 2004). This latter is particularly relevant in mate selection because the impossibility of knowing the consequences of relationship commitment beforehand (Buss, 1994) makes the confrontation with trade-offs incredibly unsettling and may engender choice resistance (Schwartz, 2004). Hence, women who can solve the adaptive problem in choosing a mate should have a comparative advantage.
We are able to examine this issue empirically because our unique cross sectional data set from the commercial online dating service, RSVP, permits a large-scale analysis of whether female participants are more likely than males to express their explicit wants and/or show actual preferences for mate characteristics. The fact that this large online population encompasses many age groups also enables exploration of how static such preferences are throughout the human life cycle. We can investigate, for example, what happens when parental investment becomes less relevant. The use of this data set also exemplifies the increasing tendency over the last two decades for behavioural science researchers to leverage the plethora of information recorded by commercial Internet dating web sites and organizations (Lee et al., 2008; Fiore et al., 2010; Whyte & Torgler, 2017a, 2017b).
The major benefit for mate choice researchers is that males and females willingly disclose large amounts of information about their personal demographics and mate preferences. Moreover, because the participation objective is the payoff of meeting a mate face-to-face, participants employ only minimal amounts of deception in their responses
(Toma et al., 2008). Yet, whereas previous research demonstrates a female preference for (and mate selection based on) education level (Hitsch et al. 2010a, 2010b, Skopek et al. 2010), psychology research into online mate choice behaviour as yet provides no large scale
59 evidence of (a distal cause for) changes in female selectiveness at different life or fertility stages. In our analysis, therefore, we leverage the openness of participant disclosure to focus on educational level, a choice characteristic that in many cultures is valued more by women than men (Buss & Barnes, 1986; Shackelford et al., 2005), possibly because of its frequent association with social status and intelligence (Correia, 2003), both common proxies for resources.
Method
Participants
Our study covers 41,936 heterosexual individuals aged 18 to 80 years, with a total of
15,194 females (M = 48.68; SD = 11.22) and 26,742 males (M = 46.93; SD = 11.79) (see
Figure 9). The dataset was generated for active members (those accessing the account at least once during the sample period) of the Australian online dating web site RSVP between
January and April 2016. The sample includes a variety of membership durations, from those who joined as far back as the RSVP launch in 1997 to those who joined as recently as April
2016 (median join date = March 2015). Large data sets such as ours are well suited for life cycle analysis in that 90% of the individuals surveyed are over 30, with only 1.29% under 25;
14.82%, 26.14%, 30.2%, and 19.95% aged between 25 and 34, 35 and 44, 45 and 54, and 55 and 64, respectively; and a mere 0.8% over 65. In addition to age, whose distribution is differentiated by sex in Figure 1, the participant online profile data also records members’ self-reported education level and number of dependent offspring. According to these data, women in the sample have higher education levels than men ( < 0.001), broken down on the web site into high school, some college, diploma, degree and post graduate. As is standard practice for online dating web sites, participants also stipulate the (educational) characteristics they prefer or are seeking in a potential partner.
60
Fig 9: Age distribution differentiated by sex
Data Collection and Ethical Practice
Data for this study were freely supplied to the researchers by RSVP as a single cross sectional data set. All RSVP members provided informed consent for the use of their demographic information and online dating behaviour for research purposes as part of their initial membership registration7. Once collected, however, all such information was de- identified to protect participant anonymity.
Measures
To provide a detailed analysis of sex differentiated educational preference in the online dating market, we utilise four different dependent variables:
1) Individual’s preference specificity differentiating between
a) Whether or not an individual preference was communicated (Fig. 10 and
11/Table 6).
7 FairfaxMedia Privacy Policy: all personal information collected may be used for “administrative, marketing (including direct marketing), promotional, planning, product/service development, quality control and research purposes, or those of our contractors or external service providers.”
61
b) The level of definitiveness (“pickiness”) based on a scale from 0 to 1 (Fig. 12).
2) Preference for homogamy or hypergamy (Fig. 13).
3) Minimum acceptable level of education in a mate (Fig. 14).
The following outlines each measure’s construction, validity and use in this study.
Preference specificity. When participants first register their RSVP account, they are asked to select as many or as few of the five educational levels as would be preferable in their ideal partner8. Individuals are classified as “picky” (or choosy) if they explicitly state a preference for education, with the selection of fewer (more) educational categories equated with higher
(lower) levels of pickiness. Because stating a preference for all options (i.e. all five levels of education) is viewed methodologically as complete indifference (Fiore et al., 2010) and effectively mirrors those who make no discrimination (choose no options) between alternatives, we code these participants as “indifferent”. In total, 70.8% (N=10,754) of females and 82% (N=21,934) of males of the whole sample did not select an option while
5.3% (N=811) and 5.8% (N=1,559) of females and males marked all five education options.
In other words, of all female participants, 23.9% state a specific preference for educational level in their preferred partner, whilst 12.2% of male registered members expressed a specific preference.
Moreover, for participants who are selective, that is, those who stated between one and four preferences (females: N = 3,629; males; N = 3,249), we also measure the extent to which individual preferences are definitive (i.e. level of pickiness) by creating a scale based on the number of categories not selected. For example, if an individual selects only one educational